You are on page 1of 215

Politecnico di Milano


Dipartimento di Ingegneria Idraulica,
Ambientale, Infrastrutture Viarie,
Rilevamento (DIIAR)

Dottorato di ricerca in Geodesia e Geomatica

A GIS embedded approach for


Free & Open Source
Hydrological Modelling

PhD dissertation
Massimiliano Cannata

Coordinatore:
Prof. Fernando Sansò
Advisor:
Prof. Maria Antonia Brovelli
Co-advisor:
Prof. Helena Mitasova


XVIII ciclo 2003-2006
Abstract

Abstract
The aim of this research is to demonstrate how watershed
management can be achieved by using Free and Open Source
Software for Geomatic (FOSS4G) through hydrological modelling.
This work was carried on in the frame of an interreg IIIA project
between Italy and Switzerland named “Sviluppo di un sistema di
gestione dei rischi idrogeologici nell’area del lago Maggiore”
(“Development of a management system for the hydro-geological
risk in the Lake Maggiore area”) partially funded by the Europe
Union.
For this purpose a new hydrological model called “HydroFOSS” was
developed.
HydroFOSS is:
 distributed – the hydrological variables are continuously
described in the space.
 Physically based – all the involved variables have a physical
meaning.
 Continuous – operates over an extended period of time,
determining flow rates and conditions during both runoff periods
and periods of no surface runoff.
 Modular – is a combination of different modules describing the
processes involved in the rainfall-runoff process.
 GIS embedded – is fully developed into a GIS system by using
the GIS's commands and library functions.
 Open Source – developed by using exclusively Free and Open
Source Software.
Due to the specific area of study (the Lake Maggiore is situated
across Italy and Switzerland, in the Alpine region) the processes
considered in the hydrological model development are the solar
radiation, the evapotranspiration, the snowmelt and accumulation,
the canopy interception, and the runoff.
The heterogeneity of the needed data, in terms of formats, topology,
coordinate systems, and time-spatial resolutions bring us to the
Abstract

development of a geodatabase that considers the time component.


For this task a methodology to handle raster series has been
developed.
Once the data have been organized in the geodatabase, the further
required step is the data processing for model input generation.
This task involves either the spatialization of numerous variables
and the validation of different data, two cases were deeply
investigated (the validation of the meteorological radar rainfall
observations, and the best interpolation technique for temperature
meteorological station observations) while, due to time restrictions,
a standard approach has been followed in other cases.
The chosen processes were then simulated by developing specific
new commands in the GIS GRASS and the overall model was setted
up by means of a general script that automatically executes all the
required operations.
Finally a link between the HydroFOSS model and the automatic
inverse calibration model UCODE-2005 (Poeter et al., 2005) was
generated and a case study application was successfully carried on.

This solution has shown how a fully open access, both in term of
cost and in term of control, to all the modelled processes and data
can be achieved by using Open Source Software and a modular
approach. Moreover the usage of a GIS allows the management of
heterogeneous data and helps models integration because of its
intrinsic data exchange, analysis and visualization capabilities.
Table of contents

TABLE OF CONTENTS
1. Introduction__________________________________________ p. 1.1
1.1 Problem Identification__________________________ p. 1.1
1.2 Objective and main steps of the research_______ p. 1.2

2. GIS and hydrological models: an overview_______ p. 2.1


2.1 Geographical Information Systems______________ p. 2.1
2.1.1 The evolution of GIS_______________________ p. 2.1
2.1.2 GIS softwares______________________________ p. 2.6
2.2 Hydrological models____________________________ p. 2.10
2.2.1 Hydrological model classification___________ p. 2.10
2.2.2 Historical review of Hydrological Models___ p. 2.15
2.2.3 Principal characteristics of some of the
currently used distributed watershed
models _____________________________________ p. 2.18
2.3 GIS for water resources_________________________ p. 2.21
2.3.1 GIS applications in hydrology_______________ p. 2.22
2.3.2 Linking GIS and hydrological model_________ p. 2.25

3. Methods_____________________________________________ p. 3.1
3.1 Chosen approaches_____________________________ p. 3.1
3.1.1 GIS-model integration level________________ p. 3.1
3.1.2 GIS package_______________________________ p. 3.2
3.1.3 Hydrological Model type___________________ p. 3.3
3.2 Rainfall-runoff conceptual model_______________ p. 3.4
3.2.1 Hydrological physical model_______________ p. 3.4
3.2.2 Hydrological logic model___________________ p. 3.5
3.3 Rainfall-runoff mathematical model____________ p. 3.7
3.3.1 Evapotranspiration________________________ p. 3.7
3.3.2 Canopy interception_______________________ p. 3.9
3.3.3 Snow melt and accumulation_______________ p. 3.10
3.3.4 Flow component____________________________ p. 3.11
Table of contents

3.4 Data structure__________________________________ p. 3.16


3.4.1 Basic input data data_______________________ p. 3.16
3.4.2 Geo-database logic model___________________ p. 3.19
3.4.3 Geo-database conceptual model____________ p. 3.21
3.4.4 Geo-database physical model________________ p. 3.23

4. System implementation____________________________ p. 4.1


4.1 Catchment maps generation____________________ p. 4.1
4.1.1 Selected approach_________________________ p. 4.1
4.1.2 Climatic map generation___________________ p. 4.2
4.1.3 Watershed characterization maps
generation _________________________________ p. 4.6
4.2 Net radiation map generation___________________ p. 4.8
4.2.1 Net shortwave radiation (Rns)______________ p. 4.8
4.2.2 Net longwave radiation (Rnl)_______________ p. 4.11
4.2.3 Global radiation (Rn)________________________ p. 4.12
4.3 Snow accumulation and melt maps generation__ p. 4.14
4.4 Potential evapotranspiration map generation___ p. 4.16
4.5 Canopy interception maps generation__________ p. 4.19
4.6 Rainfall runoff maps generation________________ p. 4.22
4.6.1 Initialization command: h.hydroFOSS.init__ p. 4.24
4.6.2 Runoff computation command:
h.hydroFOSS.runoff________________________ p. 4.27
4.7 Automatic model running_______________________ p. 4.33
4.8 Sensitivity analysis and model calibration_______ p. 4.37
4.8.1 Advanced calibration method_______________ p. 4.37
4.8.2 Calibration procedure implementation______ p. 4.39

5. Investigation of the radar rainfall estimates and of


the temperature interpolation method_________ p. 5.1
5.1 Radar rainfall estimates investigation_________ p. 5.1
5.1.1 State of the art in rainfall estimate________ p. 5.1
5.1.2 Study area________________________________ p. 5.3
Table of contents

5.1.3 Radar data________________________________ p. 5.3


5.1.4 Rainfall Data______________________________ p. 5.4
5.1.5 Preliminary analysis_______________________ p. 5.5
5.1.6 Rainfall detection analysis_________________ p. 5.6
5.1.7 Rainfall intensity analysis__________________ p. 5.7
5.1.8 Spatial correlation analysis________________ p. 5.7
5.1.9 Linear and functional correlation analysis_ p. 5.8
5.1.10 Z-R relationship adjustment______________ p. 5.9
5.1.11 Conclusions of the validation of radar rainfall
estimates________________________________ p. 5.9
5.2 Investigation of interpolation methods for
temperatures maps derivation________________ p. 5.13

6. Case study: the Verzasca basin__________________ p. 6.1


6.1 General characteristics_______________________ p. 6.1
6.2 Projection and region properties______________ p. 6.3
6.3 Topography___________________________________ p. 6.4
6.4 DTM derived maps____________________________ p. 6.5
6.4.1 Filled DEM_______________________________ p. 6.5
6.4.2 Stream, basin, flow accumulation, and flow
direction maps___________________________ p. 6.6
6.4.3 Slope map________________________________ p. 6.12
6.5 Landuse______________________________________ p. 6.13
6.6 Soiltype______________________________________ p. 6.16
6.7 Riverwidth___________________________________ p. 6.18
6.8 Model verification and running strategy______ p. 6.20
6.9 Climatological maps__________________________ p. 6.23
6.10 Watershed characterization maps___________ p. 6.29
6.11 Net radiation________________________________ p. 6.37
6.12 Snowmelt___________________________________ p. 6.40
6.13 Evapotranspiration__________________________ p. 6.45
6.14 Canopy interception_________________________ p. 6.47
6.15 Runoff calibration___________________________ p. 6.54
6.15.1 First sensitivity analysis_________________ p. 6.55
6.15.2 First model calibration__________________ p. 6.65
Table of contents

7. Conclusion and future works_____________________ p. 7.1

Acknowledgements___________________________________ p. 8.1

References____________________________________________ p. 9.1

Appendix: Meteorological Network Stations Database_ p. A.1


Chapter 1 - Introduction

1. Introduction
In the first section of this chapter the motivation that bring to the
development of this research are discussed. While in second section
will be illustrated the identified objective and main step that will be
carried on during this thesis.

1.1 Problem Identification


Recently many European countries have been affected by extreme
hazard events, therefore the associated risk for human life and
property damages rose with increased frequency.
Following a global trend, also the Alpine region shows an increasing
rate of natural hazards, that according with a study conducted in
Switzerland by WSL (Eidg. Forschungsanstalt für Wald, Schnee und
Landschaft) (Schmid et al., 2004) – based on the period 1972-2002
in the Confederation territory – are subdivided as follows:
 floods (annually 60-95% of the total);

 soil slips (annually 5-35% of the total);

 debris flows (annually 0-10% of the total);

with a mean annual “costs” calculated in 2.93 dead and 0.28 billion
of Swiss Francs (CHF).
Looking at these values and considering the expected growing
number of natural hazards due to changes in climate (different
rainfall pattern and higher intensity) and in land-cover (growing
urban areas and deforestation) it is clear that institutions have to
invest in prevention and mitigation strategies. These procedures are
particularly important and should consider transboundary
environment management strategies in order to achieve effective
improvements.
Floods have been shown to be the most frequent events thus their
risk management should be one of the major goals in the alpine
regions.
Successful floods management requires:
 the development of common sustainable flood management
strategies through land use, power plant and irrigation regulation

1.1
Chapter 1 - Introduction

all over the basins (without any administrative border limitation)


in order not to over-stress the land;
 the building or improving of flood warning systems and
emergency management tools to reduce losses of life and
property due to “unexpected” extreme hazards.
It is also evident that problems associated with water resources
cannot be solved by hydrologists alone: even if an hydrological
model could perfectly simulate the real circulation of water this is
not enough to guarantee the correct basin management (by means
of economic, social and ecological point of view).
Therefore an increasing collaboration between hydrologists,
economists, ecologists, water users, and decision makers is a key
point towards more sound policies in water resources (including
floods) management.
The collaboration needs between different disciplines and interest
groups call for more efficient sharing of data sets and methods
(which may exist but not be each other compatible).
The main candidate for grouping all this disciplines and methods is
certainly the Geo-Information science: its intrinsic interdisciplinary
and its strong linkage with the territory makes the Geographical
Information System (herein referred to as GIS) an ideal instrument
for environmental management.

1.2 Objective and main steps of the research.


Nowadays GIS is assuming an important role in Management
Information System (MIS), its ability in spatially describe and
analyse events is a key point in successful decision making
processes. It is clear that effective MIS requires both mathematical
and geographical components. While mathematical component itself
is most suited to detailed investigation of the particular
environmental processes it is not suited to collocate these processes
into their wider environmental, ecological and economic context.
Believing that GIS-based modelling allows scientists to develop
complex, interactive, and flexible applications using geospatial data

1.2
Chapter 1 - Introduction

to simulate and predict real-world events, the objective of this thesis


is to develop a GIS-based hydrological model, able to simulate the
effects of environmental policies, estimate all the hydrological
variables for further analysis and evaluate/forecast the discharges.
I also believe that such a kind of model can constitute an effective
tool for other water related hazard management systems.
The main steps of this research are:
1. Appropriate hydrological/GIS approach choice – Different kind of
hydrological model approach are available nowadays, each one
with its property and applicability. Also GIS are implemented with
different software and tools, each one with it's specific features
and programming environment.
2. System conceptual modelling – A schema describing the
considered modules and data (spatial and semantic) with the
respective structure and relationships is needed. It helps the
understanding of the relationship between the diverse inputs and
the watershed system.
3. System implementation – How the system has to be realized and
how each single block of the conceptual model has to be
developed.
4. Sensitivity analysis and model calibration procedure – While
sensitivity analysis is used to evaluate how a given model output
depends upon the input parameters (this is an important method
for checking the quality of a given model, as well as a powerful
tool for checking the robustness and reliability of its analysis),
model calibration consists in the research of the optimal (or sub-
optimal) parameters values to improve the model fit.
5. Case study application – A real case application allows to test all
the implemented procedures.

1.3
Chapter 2 – GIS and hydrological models: an overview

2. GIS and hydrological models:


an overview
2.1 Geographical Information Systems
GIS is the combination of data management with content and
locational indexing, spatial representation, and analysis techniques
to facilitate the understanding of real-world entities in their context
and interactions.
A GIS, therefore, records objects (in terms of their location and
characteristics), and provides tools (analysis and display) for their
modelling (spatial and semantic) and further modelling the effects
of characteristics and/or location changes upon the surrounding
environment.

2.1.1 The evolution of GIS


GIS technology evolved through multiple parallel but separate
applications across numerous discipline; this caused the
development of a software with interdisciplinary tools. Along its
evolution – going from the 1950s to now – we can detect three
different phases driven by the type of applications, data and
user/supplier interactions (Crain and MacDonald, 1984).
In the first phase – around the 1960s, the 1970s and the early 1980s
– GIS was a technology born as an inventory tool able to assemble,
organize and determine the extent of existent data. For this propose
it developed a defined data structure with primary data types (point,
line, area, raster) and functions for data importing, editing,
retrieving, updating, querying and reporting. The relationship
between users and suppliers was of a mere costumer-client type,
where suppliers duties where to ensure data integrity, long time
living systems (software and hardware updates), and on demand
reports and outputs. Milestones during this period are:
 1962: the first GIS system using digital data on a computer was

2.1
Chapter 2 – GIS and hydrological models: an overview

developed in Canada, by Alan Tomlinson, called the Canadian


Geographic Information System;
 1965: the development of the GBF/DIME (Geographic Base File /

Dual Independent Map Encoding) files by the U.S. Census Bureau


in the 1960s marked the large-scale adoption of digital mapping
by the government. This system led to the production of the
Census TIGER files, one of the most important socio-economic
spatial data sets in use today.
 1966: a grid-based mapping program called SYMAP, developed at

the Laboratory for Computer Graphics and Spatial Analysis at the


Harvard Graduate School of Design in, was widely distributed and
served as a model for later systems (Mark at al., 1997). These
early GIS packages were often written for specific applications
and required the mainframe computing systems available usually
in government or university settings.
 1969: Ian McHarg in his book "Design with Nature", popularized

the use of georeferenced transparent map overlays for resource


planning purposes. The Environmental Systems Research
Institute (ESRI) is founded by Jack Dangermond as a privately
held consulting firm that specialized in land use analysis projects.
Jim Meadlock establishes Intergraph Corporation (originally
called M & S Computing Inc).
 1970s, data formats begin to emerge and private vendors began

offering GIS packages. M&S Computing (later Intergraph) and


Environmental Systems Research Institute (ESRI) emerged as the
leading vendors of GIS software (Antenucci at al., 1991).
 1978, ESRI releases the first version of Arc/Info, the current

leading GIS software package. ERDAS, provider of geographic


imaging solutions, was founded by Lawrie Jordan and Bruce Rado.
 1982, ESRI's ARC/INFO® 1.0, the first commercially available

GIS software package, which ran on mainframe computers


(www.esri.com/company/about/history.html) was released.
 1982, Army Corps of Engineers Construction Engineering
Research Laboratory (CERL) started the development of the GIS
GRASS – Geographic Resources Analysis Support System – as a

2.2
Chapter 2 – GIS and hydrological models: an overview

tool for land management at military installation (Mitasova and


Neteler, 2002).
 1986, Laszlo Bardos, Andrew Dressel, John Haller, Mike Marvin

and Sean O’Sullivan founded MapInfo. ESRI's PC ARC/INFO®


1.0, the first GIS software available for the personal computer,
was released.

In the second phase – during the late 1980s and the early 1990s –
the GIS was pushed to evolve towards analysis because of the users
need of more complex retrievals and queries for hypotheses
exploring, theory confirming or data providing for research and
modelling. In this stage more functions for user interaction was
developed mainly in a graphical way by a user friendly interface
(GUI, Graphical User Interface). It gave to the user the ability to
sort, select, extract, reclassify, reproject and display data on the
basis of complex geographical, topological and statistical criteria.
The suppliers increased their knowledge on existing and growing
data analyses techniques, specific subject matters (e.g.: ecology and
hydrology), and data context issues. While data storage were still
mainly centralized the user’s access became more decentralized.
Important facts during this phase years are:
 1987, SPANS GIS (INTRA TYDAC, 1992) and the IDRISI project

(http://www.clarklabs.org/IdrisiSoftware.asp) both started.


 1988, the National Centre for Geographic Information and
Analysis (NCGIA, http://www.ncgia.ucsb.edu/) is established in
the USA.
 1991, GRASS 4.0 is released through Internet - popularity grows
with universities, business and government agencies. Vector
Library changed significantly from the library used in previous
versions.
 1992, ESRI released ArcView® 1.0, a desktop mapping system

with a graphical user interface that marked a major improvement


in usability over Arc/Info’s command-line interface
(http://www.esri.com/ company/about/history.html).
 1992, OSIRIS system corporation relesed GrassWare®: the first

2.3
Chapter 2 – GIS and hydrological models: an overview

graphical interface to the GRASS GIS software.


 1993, Steve Putz developed PARC, the first Web-based interactive

map viewer (http://mapweb.parc.xerox.com/map);


 1994, with the release of its ArcView 2.0 software, ESRI took

desktop mapping to an entirely new level: desktop GIS. The Open


GIS Consortium aiming at developing publicly available
geoprocessing specifications was founded.
 1995, ESRI released Spatial Database Engine (SDE®), an
innovative tool for storing and managing GIS data in a
commercially available database management system (DBMS).

Starting from the late 1990s GIS entered in a new era. As


computing power increased and hardware prices plummeted the
GIS became a viable technology for state and municipal planning
(Harris et al., 1993). In this third phase of evolution GIS is asked to
become a real Management Information System (MIS), and thus
able to support decision making processes.
Users look for answers to the “What if?” question, thus new
modelling and planning facilities have to be developed inside the
GIS itself. Apart from technical staff (programmers) users and
suppliers have no more distinction and are both entities of the
management system process. Pushed by the growing diffusion of
the Internet/Intranet technology the data become decentralized and
integrated control procedures are requested. In the 2000s GIS
moved to Web and become more popular for geographical
information dissemination.
This enhancement came together with growing resources
requirement , in fact:
 software development involved multiple sciences;

 data management and quality control become more complicated

due to decentralization;
 new technologies have to be tested to improve performance and

usability,
 continuity during changes should be ensured in order to let

existent application running.

2.4
Chapter 2 – GIS and hydrological models: an overview

The main facts characterizing GIS in this third phase are:


 1997, the University of Minnesota (UMN) released MapServer

1.0, an open source development environment for building


spatially-enabled Internet applications
(http://mapserver.gis.umn.edu/, 2006). ESRI released ArcView
Internet Map Server (IMS), a commercial tool for publishing GIS
data over the Internet.
 1999, GRASS 5.0 is released under GNU GPL by the GRASS

Development Team (http://grass.itc.it/, 2006). The first major


change in years, this version incorporates floating point
calculations and NULL support into GRASS raster engine.
 2001, Refraction Research released PostGIS 0.1
(http://postgis.refractions.net/, 2006), an open source "spatially
enabler" adding support for geographic objects to the PostgreSQL
object-relational database.
 2002, ESRI began offering a wide selection of GIS software

compatible with the Linux operating system. In its continuing


commitment to open systems, ArcIMS 4, ArcSDE 8.2,
MapObjects--Java Standard Edition, and ArcExplorer 4 software
are all supported on Linux.
 2005, GRASS 6.0.0 is released with new interface, vector engine,

and database support.


 2005, Google® lunches two services: the GoogleMaps®
(http://maps.google.com, 2006) that uses a new technology for
fast navigation of Web-GIS applications (AJAX, Asynchronous
JavaScript and XML) and the GoogleEarth®
(http://earth.google.com/, 2006) that adds new appeal to the map
navigation (3D fly view). They highly contribute to the diffusion of
GIS concepts and capabilities across society, through and

Figure n. 1, summarize the three phases of the GIS evolution and


the main activities as described above.

2.5
Chapter 2 – GIS and hydrological models: an overview

100
GIS evolution

75

% of applications
Inventory
INVENTORY
50
Analysis
ANALYSIS
Management 25
& data
MANAGMENT
distribution
0

ACTIVITIES: ACTIVITIES: ACTIVITIES:


•Data input & editing •Geoanalysis •Modelling & simulation
•Simple reports and maps •Statistical processing •Decision support (forecats)
•Plots and graphics •Inventory and analysis
Web-GIS
STRUCTURE: STRUCTURE: STRUCTURE:
•User/supplier is like a •Grown user/supplier •User/supplier unique
costumer/client relationship interaction

DATA: DATA: DATA:


•centralized •Mainly centralized •Decentralized
Figure 1 – GIS evolution representation after Crain and MacDonald,
1984.

Nowadays a 50% of activity are in management applications, 40% in


analyses and 10% in inventory.
Today, GIS is a user friendly software and is no more used only by
experts (geomatics): also geologists, ecologists, doctors and
economists make extensive use of this technology (user/supplier
uniqueness). It can access data held in different place, managed by
different agencies (decentralization), and stored in different format.
Applications are mostly in the hazard/impact assessment for
sustainable management (management applications) and modelling
is one of the key issues for efficient Spatial Decision Support
Systems (SDSS) development.

2.1.2 GIS software


As illustrated in the previous paragraph, a large number of GIS
software has been developed during the last decade. During their
evolution often they took inspiration each others and now as a

2.6
Chapter 2 – GIS and hydrological models: an overview

result they are quite similar in capabilities: generally they all (i)
manage numerous data format, (ii) produce customizable map
visualization, (iii) handle attributes, vector, raster and image data
types, (iv) perform general and spatial query/analysis.
We can individuate two classes of software: the commercial or
proprietary software and the Free and Open Sources Software
(herein referred to as FOSS).
The first class describes software offered for sale or license, where
the users are not allowed to see the source code. Nor are they able
to modify the code for their own use or to distribute to others. The
vendor is the only proprietary of the software and the user just
acquire a licence for use it as it is.
On the contrary FOSS denote some user freedom rights such as to
run, copy, distribute, study, change and improve the software. As
described in the Web-Site of the the Free Software Foundation
(FSF) (www.fsf.org) the word “free” refers to four kinds of user
freedom:
 The freedom to run the program, for any purpose.

 The freedom to study how the program works, and adapt it to any

needs.
 The freedom to redistribute copies.

 The freedom to improve the program, and release the


improvements to the public, so that the whole community
benefits.
For the second and fourth points access to the source code is a
precondition.
In order to guarantee this freedom, FOSS are usually licensed
under the GNU General Public License (GPL): GPL licence makes
restrictions that forbid anyone to deny these rights. Distributed
copies of a GPL licensed program, whether gratis or for a fee, must
give the recipients all the original rights of freedom; thus the
distributor must also make sure that users, too, receive or can get
the source code.
A survey currently carried on by the “GISjob” Web Site
(http://www.gisjobs.com/survey/countries.jsp, 2006) based on 27438

2.7
Chapter 2 – GIS and hydrological models: an overview

users (March 2006) shows how ESRI is the GIS market leader,
being used by the 76% of the GIS users and GRASS is the only
FOSS used (see Graph 1). This survey also shows how a part of the
GIS users take advantage of more then just one software.

Graph 1, percentage of user using different GIS software

According to graph 1 the first four GIS companies providers and the
relative software are:
 ESRI (Redlands, CA), sell ArcGis, a suite of three different

products (ArcMap, ArcCatalog and ArcTools) with a series of


extension adding features to the products. The base package is
ArcView, a desktop with visualization, query, analysis, integration,
and data automation capabilities (1'500 US$ in the 2006). Some
extension – costing 2'500 US$ each – are: 3D analyst (three-
dimensional visualization and analysis); geo-statistical analyst
(statistical tools for modelling and advanced surface generation);
network analyst (network-based spatial analysis for ArcGIS);
spatial analyst (advanced raster GIS spatial analysis).
 Autodesk (San Rafael, CA), sell Autodesk Map® 3D 2006 and

2.8
Chapter 2 – GIS and hydrological models: an overview

other products that interface with it's flagship AutoCAD software


package. This version of Autodesk Map software, can create,
maintain, and produce maps; integrate data from various sources
and formats, including Ordnance Survey MasterMap, ESRI Shape
files, MapInfo TAB and Oracle Spatial; perform data analysis; and
produce thematic maps. The 3D functionality includes the ability
to import survey point data, build Digital Terrain Models, create
contours and perform height analysis. Its price in the 2006 is
6775 US$.
 MapInfo (Troy, NY) offers MapInfo Professional version 8.0,

relatively user-friendly desktop mapping tools favoured in


demography, geology, and business analysis. In the 2006 it costs
about 1'500 US$. Its features are: analysis, geocoding, 2D and 3D
displaying, different data formats managing, geoprocessing,
spatial querying, graph producing, reprojecting, editing and
printing.
 Intergraph (Huntsville, AL) offers GeoMedia professional 6.0, its

geospatial product suite. As a high-end GIS, it provides


connectivity to industry-standard databases and file formats,
analysis tools, and tools for data creation and maintenance (price
available only by commercial sales contact).

While proprietary software grown pushed by commercial goals and


concentrated theirs efforts in developing “easy to use” software,
FOSS grew to solve problems and develop projects – often
governmental or research projects – resulting in high effective but
often not friendly software and therefore delegated to a small niche
of GIS specialists.
The most extensively used FOSS GIS is GRASS, an application with
geospatial data management, processing and analysis (vector
2D/3D, raster, volxel, images), graphics and maps production
(2D/3D), spatial modelling and visualization. GRASS is currently
used in academic and commercial settings around the world, as well
as by many governmental agencies and environmental consulting
companies.

2.9
Chapter 2 – GIS and hydrological models: an overview

Many other FOSS GIS and GIS-related software (more than 250)
can be found at the FreeGIS Project Web site (http://freegis.org).

2.2 Hydrological models


Models are a mere simplification or reduction in complexity of real
world that express our understanding of the way the phenomenon
works with sufficient precision and accuracy to allow prediction and
confident decision-making. In the specific field of surface water
resource rainfall-runoff models attempt to establish a link between
precipitation measured over the catchment and runoff (streamflow)
observed at the considered outlet. The meteorological input can be
either rainfall or snowfall depending by the air temperature.
Although precipitation and runoff can be considered as surface
waters, the transformation of the first into the second at the
catchment scale involves many hydrological processes (interception,
evapotranspiration, subsurface flow, groundwater recharge,
channel routing, etc.) of which the most important take place
underground (Beven, 2001).

2.2.1 Hydrological model classification


A possible general classification of models (Chorley and Haggett,
1967) shows four classes:
1. Natural analogues - Both historical and space analogues are used
to explain events on the basis of similar occurrence happened in
the past or somewhere else.
2. Physical scale models - Reproduce the physics of the
phenomenon with a miniaturization of reality, here is where
general behaviours, influence of variables, change in condition
are studied.
3. Mathematical - Use equation, function and statistic to describe
the physics of the real world, if a unique solution is given then
the model is labelled as deterministic else if probabilistic solution
is given then is labelled as stochastic.

2.10
Chapter 2 – GIS and hydrological models: an overview

4. Computational - A computer is used for symbolically represent


the phenomena by mean of software code.
The first three kind of models converge in the fourth one, in fact
nowadays we use all of this three types of models for develop a good
understanding of the world that is then transformed in computer
programs, able to produce fast, accurate and reusable estimations.
Computer models main importance reside in their ability to run
scenarios more efficiently assessing the environmental implications
(physical, biotic, socio-economic) that may result from the
application of policies, plans or development project.
Another prerogative of computer models are their aptitude in geo-
hazard disaster management by suggesting planning measures
(zoning, resettlement, strengthening), giving alarms (forecasting)
and assisting in post-disaster emergency (real-time monitoring).
In the field of hydrology, one of the main reason for the need of
modelling the hydrological processes is due to the lack of
measurement technique able to observe all the wanted information.
Having just a limited capability in monitoring hydrological variables
in time and space we therefore need an extrapolation of these
quantities. This also gives us the ability to forecast and to evaluate
different scenarios to assess the likely impact of future hydrological
events or land changes.
The main applications of hydrological models are intended for
managing purpose (e.g. flood mitigation; storm drainage; bridges,
sewerage and culverts design; water supply; irrigation; hydro
power; navigation; pollution control) by using model-extrapolated
information (e.g. how much water is needed; how much water can
be expected; the minimum flow; the annual flow; the floods peaks;
the floods volume; the groundwater level).
To identify the different characteristics of models a series of
attributes is used.

Metric, Conceptual and Physically Based


Following Beck (1991) a model can belong to one of these three
general classes based on the level of conceptualization of the

2.11
Chapter 2 – GIS and hydrological models: an overview

physical processes:
 Metric models - are strongly observation-oriented, and they are

constructed with little or no consideration of the features and


processes associated with the hydrological system. A metric
model omits the general laws and is in reality a representation of
data.
 Conceptual models - Describe all of the hydrological processes

component perceived to be of importance as a simplified


conceptualizations. This usually leads to a system of
interconnected stores, which are recharged and depleted by
appropriate component processes of the hydrological system.
 Physically based models - Attempt to mimic the hydrological

behaviour of a catchment by using the concepts of classical


continuum mechanics. However, has to be noted that all existing
theoretical models simplify the physical system and often include
obviously empirical components, (e.g., the conservation of
momentum equation used to describe surface flow includes an
empirical hydraulic resistance term and the Darcy equation used
in subsurface problems is an empirical equation).

Event and Continuous


Based on the time wise application it is suitable for it a model could
be defined as:
 Event models - Represent a single runoff event occurring over a

period of time ranging from about an hour to several days. The


initial conditions in the watershed for each event must be
assumed or determined by other means and supplied as input
data. The accuracy of the model output may depend on the
reliability of these initial conditions.
 Continuous models - Operate over an extended period of time,

determining flow rates and conditions during both, runoff periods


and periods of no surface runoff. Thus the model keeps a
continuous account of the basin moisture condition and therefore
determines the initial conditions applicable to runoff events. At
the beginning of the run, the initial conditions must be known or

2.12
Chapter 2 – GIS and hydrological models: an overview

assumed. However the effect of the selection of those initial


conditions decreases rapidly as the simulation advances. Most
continuous watershed models utilize three runoff components:
direct or surface runoff (overland flow), shallow surface flow
(subsurface flow) and groundwater flow, while an event model
may omit one or both of the subsurface components and also
evapotranspiration.

Complete and Partial


Based on the completeness of the represented hydrologic processes:
 Complete models - Account for all the known hydrological
processes significantly affecting runoff and maintain the water
balance, by solving the continuity equation of precipitation,
evapotranspiration and runoff: Precipitation - Actual
Evapotranspiration +/- Change in Storage = Runoff
 Partial models - Represent only a part of the overall hydrological

process. Not to solve the water balance equation decrease the


accuracy of the model and therefore it constitutes one of the most
important disadvantage of partial models compared to complete
models.

Calibrated Parameters end Measured Parameters


All the models deal with some parameters, based on the way their
values are estimated they could be defined as:
 Calibrated parameters models - Have one or more parameters

that have to be evaluated by fitting computed hydrographs to the


observed hydrographs. Calibrated parameters are usually
necessary if the watershed component has any conceptual
component models, which is true for most presently used
watershed models. Thus, with a calibrated parameter model, a
period of recorded flow is needed, usually several years, for
determining the parameter values for a particular watershed.
 Measured parameter models - Have all the parameters
determined satisfactorily from known watershed characteristics,
either by measurement or by estimation. For example, watershed

2.13
Chapter 2 – GIS and hydrological models: an overview

area and channel length can be determined from existing maps,


channel cross sections can be measured in the field, and soil
characteristics can be determined in a lab. Characteristics like
channel roughness are often estimated. A measured parameter
model can be applied to totally ungauged watersheds and is
therefore highly desirable.

General and Special Purpose


Based on the usability of the model in heterogenic basins:
 General models - Are acceptable, without modifications, to
watersheds of various types and sizes. The models have
parameters, either measured or calibrated, that adequately
represent the effects of a wide variety of watershed
characteristics. In order to achieve this, it is generally necessary
to use conceptual models which have parameters that require
calibration.
 Special purpose models - Are applicable to a particular type of

watershed in terms of topography, geology or land use, e.g. an


urban runoff model. Usually, such models can be applied to
watersheds of different sizes, as long as the characters of the
watersheds are the same.

Lumped and Distributed:


From a spatial point of view models can be subdivided in:
 Lumped models – Also known as "spatially averaged" models, do

not explicitly take into account the spatial variability of inputs,


outputs, or parameters. They are usually structured to utilize
average values of the watershed characteristics affecting runoff
volume. Averaging a certain parameter also implicitly averages
the process being represented. Because of non-linearity and
threshold values, this can lead to significant error.
 Distributed models - Include spatial variation in inputs, outputs,

and parameters. In general, the watershed area is divided into a


number of elements (e.g. cells) and runoff volumes are first
calculated separately for each element.

2.14
Chapter 2 – GIS and hydrological models: an overview

Steady and Unsteady:


Based on the time dependence of the flow equations:
 Steady models – also known as static models, consider the depth
of the accounted flows constant within a given time interval.
 Unsteady models – also known as dynamic models, consider the
depth of the accounted flows as a function of time.

2.2.2 Historical review of Hydrological Models


The development of a rainfall-runoff model for the simulation of
continuous streamflow time-series is a rather recent issue in water
research. Until the middle of the 1960s, hydrologic modelling
involved the development of models and theories of individual
components of the hydrological cycle, such as overland flow,
infiltration, evapotranspiration, interception, base flow, subsurface
flow. For example, one of the first attempts to develop a theory for
infiltration was by Green and Ampt (1911): they used simplified
principles of physics to derive a formula that is still popular for
computing the infiltration capacity rate. Later on, in 1933, Horton
developed a theory of infiltration to estimate rainfall excess and
improve hydrograph separation techniques.
The first simulation rainfall-runoff models as integration of the
different components summarized above were proposed at the very
beginning of the 1960s when modern computer appeared (Singh,
2002).
In the following 50 years an abundance of different mathematical
models has been proposed to simulate the transformation of
precipitation into stream flow (see paragraph 2.2.1). Their diversity
comes from different sources: the large number of point of view
existing between hydrologists, the wide range of physical and
climate conditions and the wide range of hydrological issues that a
rainfall-runoff model can help to solve.
The first attempt to model virtually the entire hydrologic cycle was
by the Stanford Watershed Model-SWM (now HSPF) by Crawford

2.15
Chapter 2 – GIS and hydrological models: an overview

and Linsley (1966). Immediately after another model that became


really popular was developed: HEC-1 (Hydrologic Engineering
Center 1968) by Dawdy and O’Donnel (1965). These models
represented in different “conceptual” ways the response mechanism
of the various phenomena and interconnections of the systems. In
theory, if the structural description was exact, the parameters of the
models, such as the water storage capacities, roughness coefficients
and threshold effects could be set in relation to the actual physical
quantities of the basin. Unfortunately, in many cases, due to the
large number of parameters used in the models, calibration is
needed anyway because it is impossible to have measurements for
all the parameters and sometimes it is easy to obtain a set of
unrealistic parameters.
This lack of one to one relationship between model and reality
induced Freeze and Harlan (1969) to propose a new mathematical
model based on distributed physical knowledge of the phenomena
by means of the numerical integration of the differential equations
systems describing surface flow, base flow, flow in the unsaturated
area. Unfortunately, because of the still limited computational
resources, the approach was directed only at small, almost
impermeable mountain basins.
Anyway, the Freeze and Harlan approach was later on considered
by several research instituted such as Danish Hydraulic Institute
(Denmark), the Institute of Hydrology at Wallingford (United
Kingdom) and SOGREAH (France) with the aim to develop a
physically based, distributed catchment model that would express
the phenomena by integrating the differential equations with partial
derivatives based on continuity of mass and momentum and based
on assigned boundary conditions.
The first model developed based on this philosophy is the SHE
(Système Hydrologique Européen) (Abbot et al., 1986a-b) model.
The characteristics of the SHE model will be presented in the
following paragraph.
Still of relevant importance in the 1970s is the theory of Dunne and
Black (1970) for the runoff generation. They suggested the idea of a

2.16
Chapter 2 – GIS and hydrological models: an overview

“saturation from below” (saturated excess) mechanism for runoff


production. Now on, we can have one more fundamental
characteristic for hydrological model: they can be based on the
Hortonian (hortonian excess, saturation from above) or Dunne
(saturated excess, saturation from below) approach.
The first definition of distributed model is from Freeze and Harlan
(1969). This is based on the idea that it is possible to remove some
of the subjectivity of the conceptual models using equations
describing all the surface and subsurface processes in the
catchment. In their paper, Freeze and Harlan wrote down the
equations for different surface and subsurface flow processes and
showed how they would be linked by means of common boundary
conditions into a single modelling framework.
In theory, this class of models is the best one because physical
parameters used in the model can be measured in the field and the
parameters can be defined for every element in the solution mesh.
With so many parameters the calibration can be difficult, but, at the
same time, it also impossible to measure all the parameters in the
field. Furthermore, most measurements techniques can only be
used to derive values at scales much smaller than the element mesh
used in the approximate solution. In principal, parameters should be
strongly related to the physical characteristic of the surface, soil
and rock, but in this way, we loose “cell by cell” definitions and we
build something like zones with different characteristics.
Despite these problems, the last decade saw a wide use of
distributed models, mainly thanks to the increased computer power.
The most important model included in this class is the SHE model.
The fully distributed models described here are complex and both
computationally and parametrically demanding. A separate group of
models based anyway on the fully distributed ones have been
developed: they attempt to maintain a distributed description of the
catchment responses but they do that in a simpler way, without all
the detailed process representations of the fully distributed models.
In this class we can for example include TOPMODEL, the
topographic-based model of Beven and Kirkby (1979).

2.17
Chapter 2 – GIS and hydrological models: an overview

2.2.3 Principal characteristics of some of the currently


used distributed watershed models

SHE model
The SHE model is physically-based: the hydrological processes of
water flow are modelled either by finite difference discretization of
the partial differential equations of mass, momentum, and energy
conservation, or by empirical equations derived from independent
experimental research. It uses one-dimensional finite-difference
solutions for channel reaches and the unsaturated zone in each grid
element, and two-dimensional solutions in plan for the saturated
zone and overland flow processes.
The components of the hydrological cycle are modelled as follows:
interception, by the Rutter accounting procedure;
evapotranspiration, by Penman-Monteith equation; overland and
channel flow, by simplifications of the Saint Venant equations;
unsaturated zone flow, by the one dimensional Richard equations;
saturated zone flow, by the two-dimensional Boussinesq equation;
snowmelt, by energy budget method.
In the SHE model there is the potential to have different parameters
for every grid element and within each element of the different
vertical layers. This requires the specification of thousands of
parameters.
The computational requirements of these models are extreme and,
for this reason, most current physically based models introduced
suitable simplified mathematical formulations. For example, a great
deal for the computational burden occurs in any description of a
partially saturated soil water systems because of high non-linearity
of the process. The SHE model, for example, treats unsaturated soil
water flow as a principally vertical process that links surface and
saturated subsurface hydrologic components. Such an
approximation makes physically based simulations possible for very
large catchments, although the validity of effective parameter
values that must be used with large scale catchments has been

2.18
Chapter 2 – GIS and hydrological models: an overview

questioned (Beven, 1989).


In terms of prediction, the SHE model has been proved to be
dependent on the grid scale by Refsgaard (1997).
For the SHE model impressive pre- and post-processing packages
have been developed by the different development teams.

TOPMODEL
TOPMODEL (TOPography MODEL) is a variable contributing area
conceptual rainfall-runoff model in which predominant factors
determining the formation of runoff are represented by the
topography of the basin and a negative exponential law linking the
soil transmissivity with the vertical distance from the ground level.
The spatial variability of topography is reflected in the topographic
index, k=a/tanβ, where a is the area draining through a point from
upslope and tanβ is the local slope angle (Beven, 1997). This index
is used as an index of hydrological similarity. The simplicity of the
model comes from the use of this index.
TOPMODEL represents a first attempt to combine the
computational and parametric efficiency of a lumped approach with
the link to physical theory and possibilities for more rigorous
evaluation offered by a distributed model (Beven et al., 1995).
Though a conceptual model, i.e. a model in which the physical
reality is represented as simplified, TOPMODEL is frequently
described as being “physically based”, in the sense that its
parameters can be measured directly “in situ” (Beven and Kirkby,
1979). However, this definition is optimistic, considering the
uncertainties linked in the definition of parameter values even for
physically based models. Franchini et al, 1996, studied the
TOPMODEL parametrisation with the aim to understand the
correspondence between the assumptions underpinning the model
and the physical reality, and in particular the role that topographic
index information and the nature of the soil have within the model
itself. Surprisingly, even thought that the topography plays an
important role in the formation of runoff schematization, the model

2.19
Chapter 2 – GIS and hydrological models: an overview

was found to have little sensitivity to the basin’s actual index curve,
whereas a close connection exists between the hydraulic
conductivity at the ground surface and the size of the grid of the
DEM used to calculate the index curve. In general, the hydraulic
conductivity increases with the increasing of the grid size, more
precisely, it takes values without physical meaning and this reduces
the physical credibility of the model. For another analysis of the
TOPMODEL limitations, an interesting analysis is presented by
Beven (1979).

WaSiM
WaSiM – Water balance Simulation Model – has been developed at
the Institute of Geography at ETH in Zurich (Zollmann, 2000;
Schulla, 1997, 1999).
The first version of the model is based on a TOPMODEL approach,
later on Richards equation for the unsaturated zone and an
integrated multi-layer-2D groundwater model were introduced.
The use of Richard’s equations for the description of water flow in
the unsaturated zone is the most interesting change with respect to
the old conceptual TOPMODEL approach. Furthermore the
introduction of the transport algorithms into the model makes the
model useful to simulate tracer and salt transport.
WaSiM has already been used successfully in the Swiss Alpine area,
running with a grid resolution of 100 m x 100 m. The hydropower
system has been considered externally. It is fully distributed and
considers all components of the water cycle.

TOPKAPI
Topkapi (TOPographic Kinematic Approximation and Integration)
(Todini and Ciarapica, 2001) is a distributed and physically-based
rainfall-runoff model. It is driven by a saturation excess approach
(Dunnie et al., 1978) and uses three non-linear reservoir differential
equations, for the drainage in the soil, the overland flow on
saturated or impervious soil, and the channel flow along the
drainage network, respectively. The geometry of the catchment is

2.20
Chapter 2 – GIS and hydrological models: an overview

described by a lattice of cells (the pixels of a DEM and their slope)


over which the equations are integrated to lead to a cascade of non-
linear reservoirs. The parametrisation relies on the digital thematic
maps of soil, geology and land use. Initially the Topkapi model
(Todini and Ciarapica, 2001, Liu and Todini, 2002) was structured
around five modules: evapotranspiration, snowmelt, soil water,
surface water and channel water.

2.3 GIS for water resources


As discussed in the previous paragraphs, different models have
been developed in the past, aiming at reproducing measured
variables or, more often, to predict hydrological processes in
ungauged catchments and into the future. However, the final aim of
using models is very often the improvement of decision-making
about a hydrological problem, such as flood protection, water
resources allocation, etc.
In recent years, given the necessity to evaluate the consequences of
environmental changes exacerbated by the effects of human
intervention and climate changes, researchers have turned their
attention to an integrated approach to all physical events.
Hydrological modelling is also moving in this direction, developing
physically based and spatially distributed models capable of
interacting and/or integrating with other interconnected processes.
This kind of modelling calls for the handling of a vast quantity of
data and parameters, often coming from different sources and
methodologies of measurement corresponding to different spatial
and temporal scales.
Because of its advantages of data collection, storage, management,
analysis, format-conversion, and display GIS is a very powerful and
promising tool in water resource assessment and management.

2.3.1 GIS applications in hydrology


The first kind of GIS application for hydrological purpose is related

2.21
Chapter 2 – GIS and hydrological models: an overview

to data pre- and post-processing. This type of application makes an


extensive use of the following GIS capabilities:
Geostatistic – Hydrologists use geostatistical tools to estimate the
distributed field of point-wise observations such as elevation,
precipitation, and chemical concentrations (Mitas and Mitasova,
1999). This functions are mainly used to extract input data for
hydrological models;
Map algebra and map statistic – Map algebra allows the calculation
between maps by means of arithmetic and logic operators. This is
particularly useful in parameter estimation. In fact, for instance, it
is possible to generate conductivity maps based on soil type classes
or CN maps based on landuse type (Pandey and Sahu, 2002). Also
same statistic functions for raster data (i.e. the GRASS “r.stats”
command and the ESRI “zonal statistics” command of the spatial
analyst tool) can help in determining values to be used as an input
in lumped models: they can detect for different sub-basins the
areas, the mean slope, the mean height, the percentage area for
each landuse class, and so on;
Visualization – the presentation of the results is very important
because it can strongly affect the decision-making understanding of
the modeled phenomena. Once the model outputs are imported into
the GIS, it can be used to produce 2D images and large format
maps, rendered 3D views, and animations showing the evolution of
the phenomena.
Hydrological tools – These are new hydrological procedures
developed into the GIS. Among the existing ones, we can note the
Arc Hydro Toolset, the TauDEM, the RiverTools, the hydrological
package of JGrass and the GRASS existing commands.
The Arc Hydro Toolset of the ESRI's ArcGIS suite
(www.crwr.utexas.edu/gis/archydrobook/ArcHydro.htm, 2006) is a
suite of tools which facilitate the creation, manipulation, analysis
and display of features and objects within a given hydrological data
model (ArcHydro Data Model). Some of the included operational
capabilities are for example terrain pre-processing, watershed
processing, attribute manipulation, and network utility.

2.22
Chapter 2 – GIS and hydrological models: an overview

The TauDEM (Terrain Analysis Using Digital Elevation Models) is an


interesting extendible component (plug-in) to ESRI ArcGIS
(Tarboton, 1997): a package for the analysis of digital elevation data
and mapping of channel networks and watersheds.
The RiverTools (http://www.rivertools.com/, 2006) is a user-friendly
GIS application for analysis and visualization of digital terrain,
watersheds and river networks. This software from Research
Systems Inc. provides a set of analysis tools to process and
manipulate digital terrain as well as create watershed boundaries,
flow models, stream delineations and hydrological analyses.
RiverTools is written in IDL language and is designed to work with
other GIS applications such as ESRI’s ArcView (via support for
shapefiles, BIL, FLT, GeoTIFF) and with remote sensing, image
processing systems such as ENVI (Environment for Visualizing
Images).
JGrass (http://www.hydrologis.com/jgrass.html, 2006) is a Java
based framework for the GRASS Geographic Information System. It
has been developed to generate a system independent GRASS
interface in order to facilitate its use among a larger number of GIS
professionals using a wider range of computer systems. Developed
at the International Centre for Environmental and Nuclear Sciences
of Kingston, Jamaica and at the Department of Civil and
Environmental Engineering of Trento, Italy, it is distributed with an
hydrological package by HydroloGIS (a new company that focuses
it's efforts on the environmental topic) Web Site. The hydrological
package include river extraction, flow direction delineation, fill pit
procedure, and many other hydrological commands.
Also into the GRASS GIS a series of hydrological commands exist,
some of them are:
 r.fill.dir – filters and generates a depressionless elevation map

and a flow direction map from a given elevation layer;


 r.drain – races a flow through an elevation model on a raster map

layer;
 r.slope.aspect – generates raster map layers of slope, aspect,

curvatures and partial derivatives from a raster map layer of true

2.23
Chapter 2 – GIS and hydrological models: an overview

elevation values;
 r.terraflow – flow computation for massive grids;

 r.watershed – watershed basin analysis program;

 r.water.outlet – watershed basin creation program.

2.3.2 Linking GIS and hydrological model


In order to take advantage of all the above described capabilities a
data exchanges/sharing procedure has to be generated.
This interoperability requirement between hydrological models and
Geographic Information Systems led to the establishment of links at
three different levels (Wesseling et al., 1996; Mitasova and Mitas,
2001):
1. a simple exchange of data (loose coupling);
2. an interface with GIS (tight coupling);
3. an integrated model (embedded coupling).

Loose coupling approach


This level of integration involves an external conversion program
that allows data exchanges between the two stand alone packages.
In this case the model is run independently of GIS which is used for
process and export the model input data, and import the model
results for analysis and visualization. This approach requires low
programming efforts and thus results to be the simplest of the
three. The downside is that this approach requires a big operator
work that can drives to errors .
Some examples are:
 TOPKAPI – a distributed and physically based model (Ciarapica
and Todini, 2002) whose input and output in ASCII format can be
imported and exported by ArcView and GRASS GIS;
 WaSiM – a distributed and physically based hydrological model
developed at the Swiss Federal Institute of Technology (ETH) in
Zurich has commands for importing and exporting data from and
to ArcInfo (http://iacweb.ethz.ch/staff/verbunt/Down/WaSiM.pdf,
2006)

2.24
Chapter 2 – GIS and hydrological models: an overview

 DHSVM (Distributed Hydrology Soil Vegetation Model) –


originally developed in the early 1990s by Wigmosta at the
University of Washington (Wigmosta et al, 1994), freely available
on the web
(www.hydro.washington.edu/Lettenmaier/Models/DHSVM/, 2006).
By using a suitable function (myconvert.c) it is possible to convert
an ArcInfo exported ASCII file into the binary format of the
hydrological model.

Tight coupling approach


This approach generally embeds an hydrological model into the GIS
by means of a common interface driving users in input data
processing, model running, and results analysis and presentation. In
this case the model is still run independently from the GIS, but the
common interface allows automatic procedures for data
export/import between the two systems with the effect of easier to
use model and limited human errors. This type of link is the most
diffuse because does not imply any dependence of the model from
the GIS software (possible changes in data structure, libraries or
programming tools with different versions).
Same examples are:
 SWAT (Soil & Water Assessment Tool) – developed to predict the
effect of alternative management decisions on water, sediment,
and chemical yields with reasonable accuracy for ungauged rural
basins (Neitsch et al., 2002). The SWAT/GRASS input and output
interface tools were developed by R. Srinivasan at Blackland
Research Center, Texas Agricultural Experiment Station (TAES),
Temple, Texas.
 GLEAMS (Groundwater Loading Effects of Agricultural
Management Systems) – is a field-scale model used to evaluate
the effects of management practices on movement of chemicals in
the plant root zone (Knisel, 1993).
 WEPP (Water Erosion Prediction Project) – a process-based,
distributed parameter, continuous simulation, erosion prediction
model firstly used with GRASS by Sabivi et al. (1995).

2.25
Chapter 2 – GIS and hydrological models: an overview

 GeoWEPP (Geo-spatial interface for the Water Erosion Prediction

Project) – an enhanced version of the WEPP model interfaced with


ArcView 3.x and with ArcGIS 9 (relese of May 2006) (Renschler,
2003). It allows the import of DEMs, the channel and watershed
delineation, the model input parameter setting and the model
running.

Embedded coupling
Into this methodology of integration two opposite subtypes of
approach can be recognized. The first one involves the development
of GIS-like capabilities into the hydrological model: this leads to big
programming efforts and usually limited GIS functions, the
advantage is that the development is with no data structure
constrains (it is GIS software independent).
Some examples are:
 RiverCAD – is a hydraulic model with CAD embedded capabilities
with support for HEC-2 e HEC-RAS.
 MIKE-SHE – a fully modular system for mathematical description

of the land phase of the hydrological cycle derived from the SHE
model (Abbot at al., 1986a, 1986b); it integrates modelling of
surface water and ground water and has a sophisticated GUI
(Graphical User Interface) for data manipulating and
visualization.

The other approach embeds hydrological model routines into the


GIS software: this leads to the development of the model by using
GIS libraries, and thus taking full advantage of all GIS capabilities
and avoiding any needs of data import/export and duplication (data
in the GIS format and data in the model format).
 TOPMODEL – a widely used semi-distributed topographically
based model, (Beven and Kirkby, 1979) of which a GRASS version
exists (r.topmodel).
 CASC2D – is a fully-unsteady, physically-based, distributed-
parameter, raster (square-grid), two-dimensional, infiltration-
excess (Hortonian) water surface model for simulating the

2.26
Chapter 2 – GIS and hydrological models: an overview

hydrologic response of a watersheds subject to an input rainfall


field (Ogden, 1998). In addition to a loose coupling level of
integration with ArcInfo, a completely GRASS integrated version
of the model is available (r.hydro.CASC2D).
 GSSHA - is a physically based, distributed-parameter, structured

grid, hydrologic model that simulates the hydrologic response of a


watershed subject to given hydro-meteorological inputs
(http://chl.erdc.usace.army.mil/software/gssha/Primer_20/gssha_p
rimer_20.htm, 2006). It is an evolution of the CASC2d model. It is
fully embedded in the Watershed Modeling System (WMS), a GIS-
based data processing framework for watershed hydrology and
hydraulics modelling developed by the Scientific Software Group
company
(http://www.scisoftware.com/products/wms_overview/wms_overvi
ew.html, 2006).
 LISFLOOD – a physically based river basin model (De Roo et al.,
2000), written using the PCRaster GIS environment; the PCRaster
language is a computer language for construction of spatio-
temporal environmental models and supports immediate pre- or
post-modelling visualisation of spatio-temporal data (Van
Deursen, 1995).
 ANSWERS (Areal Nonpoint Source Watershed Environmental
Response Simulation) – is an event oriented, distributed
parameter model that was developed to simulate the behaviour of
watersheds having agriculture as their primary use (Beasley at
al., 1980). It is present into the GRASS GIS with the command
r.answers.

2.27
Chapter 3 - Methods

3. Methods
Recent development of GIS and remote sensing technology make
possible to capture and manage a vast amount of spatial data
relative to hydrological parameters and variables. Linking GIS and
hydrological models is therefore of rapidly increasing importance to
test and validate these data in watershed management.

3.1 Chosen approaches


In this paragraph I present the choices of the GIS-model integration
level approach, the GIS package and the hydrological model type.
Motivation and advantages are also discussed.

3.1.1 GIS-model integration level


As discussed in paragraph 2.3.5 three integration level between
models and GIS exist.
With the present research I intend to develop a system able to be
used as a Watershed Management Information System (WMIS). This
involves the ability to link geospatial data (stored into GISs) with a
model (describing the evolution or state of phenomena, in this case
the water flow) and the GIS's analysis tools (to evaluate the
phenomena with respect to a further analysis of the relationship
with other factors like for example landuse changes).
The GIS embedded approach allows the development of such a
system in a single environment avoiding the time consuming
operations like data import/export and format conversion that
otherwise are needed.
Furthermore this approach allows an efficient and not redundant
development: existing algorithms (i.e interpolation techniques, map
algebra, basin extraction) and existing optimized libraries (i.e.
compressed format files, optimized data access and writing) can be
used avoiding code duplication.

3.1
Chapter 3 - Methods

3.1.2 GIS package


Among the existing GIS packages described in the literature review
(chapter 2), GRASS is the software chosen for this research.
A short description of this GIS software given by Dr. Helena
Mitasova in one of the GRASS mailing lists is: “Geographic
Resources Analysis Support System. Commonly referred to as
GRASS, this is a Geographic Information System (GIS) used for
geospatial data management and analysis, image processing,
graphics/maps production, spatial modelling, and visualization.
GRASS is currently used in academic and commercial settings
around the world, as well as many governmental agencies and
environmental consulting companies”. Further information about
the GRASS GIS can be found on the official GRASS web page
http://grass.itc.it/ (2006).
Five relevant points motivate this choiche:
 data format – it handles all the three existing data structures:

raster, vector and database. It is able to use a large number of


format inside each data structure by using OGR-GDAL library (for
raster and vector format conversions) and PostgreSQL, DBF,
SQLite, MySQL, and ODBC drivers (for database connection).
 Code access – it is a public domain free and open source GIS.

 Documentation and support – t is supperted by an extensive

documentation with programming manuals, commands manuals,


books, tutorials, and a huge community with numerous users
groups, mailing lists, and a Wiki site (a collaborative Web site
including the perpetual collective work of many authors: a Wiki
allows anyone to edit, delete or modify content that has been
placed on the Web site using a browser interface)
 Programming language – it is written in standard ANSI C, a

language robust, fast, and widely used in modelling.


 Specific existing functions – it includes a large number of
hydrological modules that are used in the development of this
research.

3.2
Chapter 3 - Methods

3.1.3 Hydrological Model type


The chosen model type is: continuous, distributed, physically based,
and modular.
The continuous and distributed approach has been selected because
a Watershed Management Information System (WMIS) should be
able to estimate the state of the hydrological quantities over time
and space. For example it is useful to understand the influence of
landuse changes, to implement a flood warning system, to monitor
the catchment status, to predict the state of ungauged catchment,
and to estimate the movement of pollutants and sediments.
The physically based approach has been selected because it
attempts to link catchment behaviour with measurable physical
landscape properties (i.e. soil moisture); this can be a valuable
characteristic in coupling with other interconnected models (i.e.
landslide hazards model) and in model validation.
I used a modularity approach(simulation of the different processes
by using separate functions) because:
 it can be helpful in future improvements or changes of the code;

 it allows a full control of the rainfall-runoff process in terms of

freedom, in combining and choosing the desired processes;


 it permits the use of single modules for whatever purpose (i.e.

evaporation for irrigation planning);

3.3
Chapter 3 - Methods

3.2 Rainfall-runoff conceptual model


In this paragraph the physical model, the logic model and the
mathematical model of the rainfall-runoff process are described.

3.2.1 Hydrological physical model


In the rainfall-runoff modelling the aim is to simulate the water
movement of a sub-system of the water cycle that is related to the
watershed area (herein referred to as watershed system). That sub-
system is therefore linked to the global water circulation and
constitutes a process with exchanges of mass and energy. As
represented in figure 3.2.1, the watershed system includes a series
of sub-processes.

Figure 3.1 – watershed physical model

At first, the precipitation can be solid (snow) or liquid (rain),


depending on the air temperature. It can fall over the canopy and a
part of it can be intercepted while the rest of the rain reaches the
ground. Then the energy of the sun induces the water
evapotranspiration and/or the snow melting. Later on the water on
the ground flows following two ways: along the surface (faster
process) or along the deeper layers (slower process).
The first way involves the flows over and into the first thin upper

3.4
Chapter 3 - Methods

soil layer; the second way involves the percolation and the flows
into the deeper soil layer.

3.2.2 Hydrological logic model


Schematizing the physical model of the watershed system the logic
model can be derived (see figure 3.2).

ET
T E
P P P P
Se

vegetation snow

Pd Sm
surface
Qs

channel
Qi network
upper soil

Qb
deeper soil

Figure 3.2 – watershed logic model (P=precipitation, Pd=drained


precipitation, Sm=snowmelt, Qs=surface discharge, Qi=subsurface
discharges, Qb=groundwater discharges, E=evaporation,
T=transpiration, ET=evapotranspiration, Se=solar energy)

In this schematization of the watershed system the precipitation P


(mass input) and the solar energy Se (energy input) are the two
inputs, while the river discharge Q (mass output) and the
evapotranspiration (mass output) are the two outputs.
The water during its transformation from the input (rainfall) to the
output (runoff) pass through different stages that accumulate,
delay, route and lose part of the mass.
The processes that mainly describe these stages are:
 the vegetation module – it mimics the canopy interception

process causing mass losses;


 the snow module – it mimics the snow accumulation and melt

process causing accumulation and therefore delay;


 the surface module – it mimics the routing of water over the

3.5
Chapter 3 - Methods

surface causing low delay;


 the upper soil module – it mimics the routing of water into the

first thick soil layer causing medium delay;


 the deeper soil module – it mimics the routing of water into the

deep soil causing high delay;


 the channel module – it mimics the routing of the water in the

channels causing very low delay and output generation.


 The solar energy affects different stages and it is used to

evaporate the water intercepted from the canopy; to transform


the accumulated snow to water; and to evaporate the water that
resides in the upper soil.
For model implementation all these modules have to be formally
defined by mathematical equation that represents the single
process: the next section will describe the equations governing the
different processes.

3.6
Chapter 3 - Methods

3.3 Rainfall-runoff mathematical model


In the following paragraphs the chosen approach for each
component of the logic model is described and the relative
mathematical formulation is reported.
Looking at the existing solutions I used different formulation found
in literature that shown proven accuracy. Following this idea the
SHE approach was used to simulate the snow melt, the Penman-
Monteith approach to simulate the evapotranspiration, the TopKapi
approach to simulate the soil flux component, and the Rutter
approach to simulate the canopy interception

3.3.1 Evapotranspiration
This physical process is controlled by both the energy availability on
the evaporating surface and the capability of which the water
vapour to be diffused in the atmosphere. Methods, that differ in the
specific data availability (temperature, global radiation, wind
velocity, etc.) and the temporal resolution (hourly up to several
days), have been developed and can be found in literature
(Shuttleworth, 1992; Xu and Chen, 2005).
The approach used in this work is the hourly Penman-Monteith
method as presented in Allen et al. (1998) for land surfaces and the
Penman method (Penman, 1948) for water surfaces.
This choice is due to:
 no calibration needs – unlike other methods, neither a local

calibration nor the definition of the wind field on the basis of the
measurements are needed: it allows to comparing data from
different zones;
 it is physically based – it explicitly contains the parameters

describing both physiological and aerodynamic components;


 it is recommended by FAO – the results of the studies by many

authors summarized in the document of the experts of FAO (Allen


at al., 1998), suggest to use the Penman-Monteith method for the
determination of the potential evapotranspiration.

3.7
Chapter 3 - Methods

Being:

λ = latent heat of vaporisation [MJ·Kg-1];


∆ = slope of vapour pressure curve [kPa·°C-1];
γ = psychrometrics constant [KPa·°C-1];
γ* = modified psychrometric constant (rs and ra dependent);

ρ = atmospheric density [Kg·m-3];


cp = specific heat of moist air [KJ·Kg-1·°C-1];
rs = bulk surface resistance [s·m-1];
ra = aerodynamic resistance [s·m-1];
es = mean saturation vapour pressure [KPa];
ea = actual vapour pressure [KPa];
Rn = net radiation [W·m-2];
G = soil heat flux [MJ·m-2·d-1];
u2 = wind speed at 2 meters height [km·h-1];
ETp = potential evpotranspiration [mm·h-1];
ETaero= aerodynamic term of the evapotranspiration [mm·h-1];
Etrad = radiative term of the evapotranspiration [mm·h-1];
f(u) = mass transfer coefficient [mm·mb-1 ]

The applied equations for Penman-Monteith are:


ETp = ETrad + ETaero;
ETaero = (0.001/λ)·[1/(∆ + γ∗)]·(ρ·cp/ra)·(ea – es); [eq. 3.1]
ETrad = [∆/(∆ + γ)]·[(Rn – G)/λ ·10 ];
6

and for Penman are:

ETp = [ (Rn · λ · ∆) + (γ · ETaero) ] / (∆ + γ);


ETaero = f(u) · (ea – es); [eq. 3.2]
f(u) = (0.35/24) · (0.5+(0.621375·u2/100)) · 7.500638

where f(u) is expressed following the AWSET software approach


(www.silsoe.cranfield.ac.uk/iwe/_private/software/awset/awset3.pdf,
2006).

3.8
Chapter 3 - Methods

3.3.2 Canopy interception


Canopy interception plays an important role in determining the
amount of rainfall reaching the soil surface. The amount of rainfall
intercepted by a forest canopy depends on the storm size, intensity,
and duration, the rainfall frequency, the forest structure, and the
tree species, age, and density. Depending on all these conditions,
canopy interception can account for 15 to 35% of annual rainfall,
although the influence during intense storm events is obviously
reduced (Blake, 1975).
The chosen method is the one presented in Zengh et al. (2000), a
Rutter approach (Shuttleworth, 1988) modified to take into account
the temporal variability of rain. This approach has been selected
because it is physically based and permits to estimate different
interception losses for drizzles (higher) and intensive showers
(lower).
The canopy interception is modelled by means of a storage
described by a maximum storage capacity (Cm [mm]), filled by
precipitation (P [mm]) and dried by evaporation (IL [mm]) and
drainage (D [mm]), both in function of the current storage level (Ca
[mm]).
The water balance for the canopy storage is evaluated by:

∂ Ca [eq. 3.3]
= P − IL − D
∂t

The intercepted water exceeding Cm is assumed to drain


instantaneously to the ground and therefore the drainage D is
modelled by the following equation:

 ∞ Ca > Cm [eq. 3.4]


D= 
 0 Ca ≤ Cm

and the IL (called interception loss) is:

3.9
Chapter 3 - Methods

C 
IL =  a  ⋅ ETp [eq. 3.5]
 Cm 

with ETp = evaporation from a water surface.

3.3.3 Snow melt and accumulation


As shown in Harr (1981), snow melt in mountainous region is an
important factor in surface water balance.
Despite the numerous complex models proposed in literature (i.e.
Morris, 1991; Rango and Martinec, 1995; Tarboton & Luce, 1996),
the World Meteorological Organization (WMO, 1986) shows how no
clear advantage in using high complex model can be detected.
Thus the SHE model approach (Abbott at al., 1986a,b) has been
chosen because it is quite simple but still physically based.
The snow pack dynamic is driven with a simplified energy budget.
By using the hypothesis of zero snowmelt the water equivalent mass
(Z*) and the energy (E*) are estimated with the equations 3.6:

Z * ( t + ∆ t ) = Z ( t ) + Ps
E * ( t + ∆ t ) = E ( t ) + Rad + Qp; [eq. 3.6]

where: Ps is the estimated solid precipitation, Rad is the available


net solar radiation, and Qp is the energy advected by precipitation
(solid and liquid) into the snow.
If the total available energy is bigger than that required to maintain
the total mass in a solid phase at the temperature T0 of 0°C, then the
snowmelt, the updated energy (E) and mass content (Z) of the
snowpack are calculated by means of the equations 3.7:

3.10
Chapter 3 - Methods

E * ( t + ∆ t ) − C si ⋅ T0 ⋅ Z * ( t + ∆ t )
Sm =
C lf
[eq. 3.7]
Z ( t + ∆ t ) = Z * ( t + ∆ t ) − Sm
E ( t + ∆ t ) = E * ( t + ∆ t ) - ( C si ⋅ T0 + C lf ) ⋅ Sm;

where: Csi is the specific heat of ice ,Clf is the specific latent heat of
fusion of water, and Sm is the snowmelt.
Otherwise no snowmelt occurs and the snowpack energy content
remains available for further steps.

3.3.4 Flow component


As described in the logical model of the rainfall-runoff process the
water that reaches the ground, once depleted of the
evapotranspiration ratio, is available for the runoff.
The runoff may occur on the ground surface (surface flow), in the
first thick soil layer (subsurface flow), in the deeper soil
(groundwater flow), in the rivers (channel flow), through the lakes
(lake), and through the artificial basins (dam).
As already mentioned the Topkapi approach is limited in parameters
number and it is scale independent .
For the sake of simplicity, in this first version, only the three major
flow component involved in floods generation are considered: the
subsurface flow, the surface flow, and the channel flow.

Subsurface flow
There are five fundamental basic assumptions for the description of
this component:
1. precipitation is assumed to be constant over the integration
domain (namely the single cell), by means of suitable
interpolation of the local rainfall data, such as Thiessen polygons
techniques, Block Kriging or others.
2. all the precipitation infiltrates into the soil, unless the soil is
already saturated; this is equivalent to adopting, at the pixel
scale, the saturation excess or Dunne mechanism (Dunne, 1978),

3.11
Chapter 3 - Methods

instead of the possible activation of the Hortonian mechanism


due to infiltration excess;
3. the slope of the water table is assumed to be equal to the slope of
the ground, unless the latter is very small (less than 0.01%); this
is the fundamental assumption of the approximation of the
kinematic wave in the De Saint Venant equations, and it implies
the use of a kinematic wave propagation model with regard to
horizontal flow, or drainage, in the unsaturated area (Henderson
and Wooding, 1964; Beven, 1981, 1982);
4. the integral of the hydraulic conductivity over the vertical in the
unsaturated zone, an extension of the concept of transmissivity,
together with the horizontal flow, can be reasonably expressed as
a function of the total water content of the soil, i.e. on the basis
of the integral of the water content profile in the vertical
direction ;
5. saturated hydraulic conductivity is constant with depth in a
surface soil layer but much larger than that of deeper layers; this
is the basis for the vertical aggregation of the transmissivity, and
therefore of the horizontal flow.
The kinematic wave formulation for sub-surface flow is then based
on the approximation on horizontal transmissivity, the continuity
and momentum equations for soil water and is expressed as follows:

~
 ∂Θ ∂q
(
 s ϑ − ϑ r ) L + = p
 ∂t ∂x [eq. 3.8]
 q = tan ( β ) k LΘ~ α
 s

where x is the main direction of flow along a cell (dimension of the


cell) [m], t is the time [s], q is the flow in the soil due to drainage
[m3/s], corresponding to a discharge per unit of width, p is the
intensity of precipitation [mm/h], θs is the saturated soil water
content [m3/m3], θr is the residual soil water content [m3/m3], β is the
soil slope [deg], ks is the soil conductivity [m/s], L is the soil
thickness [m], Ố is the mean reduced water content along the

3.12
Chapter 3 - Methods

vertical profile [-], and α is a soil type dependent parameter [-].


The model is written in just one direction since it is assumed that
the flow along the slopes is characterised by a preferential
direction, which can be described as the direction of maximum
slope.
From the combination of the two equations (3.8), considering the
problem in terms of actual total water content in the soil, η=(θs-θr)L
Θ, along the vertical profile, and making the following substitution:

Lk tan ( β )
C= s
[eq. 3.9]
( )
α α
ϑ −ϑ L
s r

we obtain the following kinematic equation:

∂η ∂q ∂ ( Cη α )
= p− = p− [eq. 3.10]
∂t ∂ x ∂ x

Where, from the physical point of view, the term C representing a


local conductivity coefficient depends on soil parameters and is
directly proportional to physically meaningful quantities such as
hydraulic conductivity and slope, and inversely proportional to the
storage capacity.

Surface and channel flow


Remembering the assumption for which all the precipitation
infiltrates into the ground, the input to the surface water model is
the precipitation excess resulting from the saturation of the soil
layer. Furthermore, the flow in the soil can exfiltrate, thus also
feeds the overland flow.
The description of overland flow is similar to the soil component,
according to the kinematic approach (Wooding, 1965), in which the
momentum equation is approximated by means of the Manning's
formula. For a general cell, the kinematic wave approximations of

3.13
Chapter 3 - Methods

overland flow and channel flow are described in equations 3.11 and
3.12.

 ∂ ho ∂ qo
 ∂ t = ro − ∂ x
 [eq. 3.11]
 1 5
 q = 1 (tan β ) 2 h 3 = C h α o
 o no o o o

where h0 is the water depth over ground surface in m, r0 is the


saturation excess resulting from the solution of the soil water
balance equation, either as the precipitation excess or the ex-
filtration from the soil in absence of rainfall, in m/s, n0 is the
Manning friction coefficients for the surface roughness in m-1/3s-1,
Co=(tanβ)1/2/no is the coefficient of the Manning formula for overland
flow, αο=5/3 is the exponent which derives from the use of Manning
formula.

 ∂ hc ∂ qc
 ∂ t = rc − ∂ x
 [eq. 3.12]
 1 5
 q = 1 s 2h 3 = C h α c
 c nc 0 c c c

where hc is the water depth in the channel reach in m, rc is the


lateral drainage input, including the overland runoff reaching the
channel reach and the soil drainage reaching the channel reach in
m/s, so is the bed slope, assumed to be equal to the ground surface
slope tan(β), nc is the Manning friction coefficients for the channel
roughness in m-1/3s-1 , Cc=so1/2/nc is the coefficient relevant to the
Manning formula for channel flow, αc=5/3 is the exponent which
derives from the use of Manning formula.

3.14
Chapter 3 - Methods

3.4 Data structure


According to the formulation described in the paragraph 3.3
different input data are required for each module.
In the paragraph 3.4.1 I will just shortly list them in order to
understand how to design the logic model of the geo-database
(described in the paragraph 3.4.2) that will be used by the
hydrological model.

3.4.1 Basic input data


The basic data, from which all the other needed input data can be
derived, can be classified in three different groups: territorial data,
meteorological data, and soil-land semantic data (i.e. hydrological
information like literature Manning coefficient values for different
landuses, vegetation parameter like the canopy height at given
reference dates) herein referred to as “watershed characterizing
data”.

Territorial data
They are the digital elevation model (DTM), the landuse, and the
soiltype raster maps. These three maps will be used to derive other
informations such as the water flow directions or the soil
conductivities.

Meteorological data
This data set includes maps of air temperature, air humidity, wind
speed, and rain intensity.
These information are usually acquired by using a network of
meteorological stations and the relative maps are generated by
means of spatial interpolation of the point-wise observations. Some
of these information can also be extrapolated by remote sensing
observation: such as, for instance, the thermal infrared satellites
sensors for temperature like the MODIS Land Surface Temperature
product (http://modis-land.gsfc.nasa.gov/temp.htm, March 8, 2006).

3.15
Chapter 3 - Methods

Watershed characterizing data


These data set describe the basin through indexes representing the
characteristics of the system.
These indexes are:
 Leaf Area Index (LAI) – the one sided green leaf area for unit

ground area in broadleaf canopies. It is generally used to predict


the photosynthetical primary production of vegetation that drives
the transpiration process: a higher LAI corresponds to a higher
leaf mass and therefore a higher respiration rate of the crop. It's
values, that can be directly observed, are generally derived from
the literature (obtained from a statistically significant sample of
plants from a crop) as a function of the vegetation type (landuse
category) and of the crop growth (function of time and quota).
Alternative indirect methods to obtain the LAI values are
algorithms that derive this index from the satellite data, one
example is the MODIS “Leaf Area Index/FPAR 8-Day L4 Global
1km” product (Knyazikhin et al., 1999).
 Vegetation cover – the percentage of a unit area covered by

vegetation; this index is used to detect the areas influenced by


vegetation dependent processes, like evapotranspiration or
rainfall interception. It's values can be either be extracted by
literature as a function of the vegetation type (landuse category)
and of the crop growth (function of time and quota) or by
algorithms that computes it from the normalized differential
vegetation index (NDVI) derived from satellite data
(www.essc.psu.edu/frac_veg/methodology/, 2006).
 Vegetation height – the mean height of the vegetation for unit
areas; this index is used to derive the “zero plane displacement
height” (the level of zero wind) in vegetated areas that allows the
determination of the aerodynamic resistance value (the transfer
of heat and water vapour from the evaporating surface into the
air above the canopy). It's values can be derived from literature
values as a function of the vegetation type (landuse category) and
of the crop growth (function of time quota) or from LIDAR (Light
Detection and Ranging) observations (St-Onge and Achaichia,

3.16
Chapter 3 - Methods

2001).
 Albedo – a (unitless) measure of reflectivity of a surface or body.

It is the ratio of total electromagnetic radiation (EM radiation)


reflected to the total amount incident upon it, thus it represents
the fraction of solar energy (shortwave radiation) reflected back
into the space; It's values can be derived either from literature
values as a function of the land typology (landuse category) or
from satellite observations such as the MODIS “BRDF/Albedo
Product” (Strahaler et al., 1999)
 Linke – an index of the atmospheric absorption and scattering of

the solar radiation under clear skies. It describes the optical


thickness of the atmosphere due to both the absorption by the
water vapor and the absorption and scattering by the aerosol
particles relative to a dry and clean atmosphere. It summarizes
the turbidity of the atmosphere, and hence the attenuation of the
direct beam solar radiation (WMO, 1981; Kasten, 1996). The
larger TL, the larger the attenuation of the radiation by the clear
sky atmosphere. It's values can be derived from literature as a
function of lanuses and months, or derived from the SoDa
“Service for Knowledge in Solar Radiation: Data, Databases,
Applications, Education (SoDa: Solar Data)” databases
(http://www.soda-is.com/eng/services/climat_eng.html, 2006).
 Hydrological indexes (Manning, soil water contents,
transmissivity exponent, horizontal permeability, and soil
thickness) – they characterize from an hydrological point of view
the soil. Theirs values can be measured “in situ” or derived from
literature as a function of the soiltype category. Even if these
indexes are physically based, their values can be different from
the directly observed, in fact when applied for gridded areas, they
assume a mean value over the cell, thus a calibration of these
values is often suggested. Further informations on standard
values can be found in section 6.8 dedicated to the case study.
As described above most of these data at a given time (date and
hour) can be derived combining landuse and/or soiltype and/or
elevation maps in conjunction with literature values of the

3.17
Chapter 3 - Methods

parameters (both monthly or seasonally known).


Other common sources for these parameters are remotely sensed
data (e.g. Terra & Aqua MODIS Data Products,
http://edcdaac.usgs.gov/dataproducts.asp, January 09, 2006). Those
sources have the advantage of measuring, with a correct spatial
distribution, the current values and therefore considering all the
anomalies that can occurs (like as an example the delay in crops
growing due to a particularly dry season). The drawback is that they
aren't direct measures and often their validity have still to be
proven.

Once the needed basic data has been detected, the Geo-database
logic model can be modelled.

3.4.2 Geo-database logic model


The GRASS geo-database is structured in three subsequent levels:
 GISDBASE – the directory housing all the geo-database;

 LOCATION – the directory housing a spatial domain;

 MAPSET – the directory housing a user project;

In each LOCATION a special MAPSET called PERMANENT is


automatically generated; this is where the maps available to all
users should be placed. Every user will store his own map in his
own MAPSET, where other users has no access in writing mode.
Into each MAPSET, there are two main folder sets, one related to
the raster maps, and the other related to the vectors maps
geometry. All the attributes can be stored into an external database
table or into an internal dbase (dbf ) file.
From a logic point of view, hydrological modelling involves the
usage of three kind of data sets: point-wise (climatic station
observations), raster (distributed observations, extrapolations, or
simulations), and attribute (literature parameters and general
informations).
The adopted logic model of the data is represented in figure 3.3.

3.18
Chapter 3 - Methods

PostgreSQL GRASS DB
vector
vector
spatial
attributes
component

GRASS DB
Database

literature
raster
values..

Gis modules:
visualization, analysis,..

GRASS CORE
Figure 3.3 – logic data structure.

3.4.3 Geo-database conceptual model


The conceptual model of the geo-database is represented in figure
3.4 where three different areas can be detected: point data series,
raster series and static rasters.

Point data series:


Each climatic station, owned by an organization, collects, at
different time interval, observations of various parameter types.
Maintenance informations are also considered by storing involved
people and their operations.

Raster series:
Some distributed values are dynamic and therefore have different
representation for different time: for example the rainfall field

3.19
Chapter 3 - Methods

typically change over space and over time.

Static rasters
Some distributed values are “static” (DEM, landuse, soiltype,etc.)
and some of their classes could be related to monthly or seasonal
parameters.
From a spatial point of view the climatic stations represented with
the point primitive and the distributed values (series and static) are
represented with rasters.

Figure 3.4 – Logic data model, red squares indicate raster data and red
point indicate point data

3.20
Chapter 3 - Methods

3.4.4 Geo-database physical model


The final aim of an hydrological model is to represent water fluxes
over a spatio-temporal space: this leads to the establishment of
procedures to correctly represent this spatial and temporal
dimensions.

Point data series


Vector data in GRASS are represented by a geometry native file
linked with a table of attributes trough a unique feature
identification number (cat) present both in the file and in the table.
The spatial component of point data series (measurements stations)
is fixed and therefore it should be stored as a static geometry native
file in GRASS while their attributes (observed values) at a given
time should be dynamically linked to the geometry.
Unfortunately at this time this option can not be realized: in fact
GRASS data structure allows to establish a link only between a
geometry file and an attribute table, but not with a selection of it.
This lack leads, for each considered time instant, to the generation
of a new vector files which will be used for the interpolation process
and then erased: obviously this is not the optimal procedure and a
modification of the GRASS vector structure is desirable in order to
handle time series data.

Raster series
Also the distributed values change and the surface describing
variables is different time to time (i.e. temperature field).
Conceptually a three dimensional (2D in space + 1D in time)
distributed information could be well represented by voxels; the
drawback of this possible choice is that the file dimension could
become huge (around 8700 levels a years are needed) and
numerous operations of cutting (computationally expensive), aimed
to extract information at a time instant, could be required.
Usually in hydrological models time is represented by consecutive
steps: like photograms in a movie, ASCII files are used to store the
status of the system at consecutive instants.

3.21
Chapter 3 - Methods

Transferring this solution into GISs, series of rasters should be used


by find out a method to relate the raster files with the respective
time instants.
Even if GRASS handle time information of a layer (vector or raster)
by writing/reading a timestamp in a specific file (Neteler M., 2005) I
decided to adopt a more direct and fast approach by including the
time information into the raster name suffix.
The suffix is composed by a 12 digit number representing:
 year (1st to 4th digit);
 month (5th to 6th digit);
 day (7th to 8th digit);
 hour (9th to 10th digit);
 minute (11th to 12th digit).
Therefore the raster representing the rainfall on the 12th of August
2005 at the 17:00 will be named “rain.20050812170”.

Static maps
Not dynamic features can easily be handled by common GIS vector
or raster files and the monthly or seasonally known parameters
values can be stored in database tables. The relationship between
rasters and parameters can be done by using raster category values
as a key field of the attributes tables.

3.22
Chapter 4 - System implementation

4. System implementation
In this chapter I present the solution adopted and the modules
development that bring to the model realization.
The programming languages used are the GRASS native languages:
the C ANSI for implementing new modules and the Bash Shell for
scripting.
The resulting model, aggregation of different procedures and
modules called in the right sequence, will be herein referred to as
HydroFOSS (Hydrological Free and Open Source Software).

4.1 Catchment maps generation


A distribute model need distribute inputs.
Distribute inputs can be derived from distributed observations (i.e.
satellite images) or from spazialization of point wise observations
(i.e. climatic stations) or literature values (i.e. leaf area index).
Spazialization of point wise observations and literature values can
be achieved applying different methods.

4.1.1 Selected approach


In the present work I adopted the classical approach (the most
diffuse and tested approach in hydrology modelling) that uses point
wise observations mainly regionalized by means of Thiessen
tasselletion and literature values distributed on the basis of landuse
and soiltype classes.
I performed a more detailed investigation, presented in the next
chapter 5, only for the analysis of distributed rainfall estimates
obtained from the meteorological radar, and for the investigation of
the best method for temperature interpolation.
These analyses reveal that nowadays radar rainfall estimates in
alpine regions seem to be still too inaccurate to be used with
confidence, and that the best method for the temperature

4.1
Chapter 4 - System implementation

interpolation is a simple linear gradient.


Further studies should be carried on in future works.
A general consideration is that even if the catchment distributed
maps can be both derived from remote sensing, this technology
presents at the moment the following limitation:
 it is not jet commonly validated, being a quite new object in
hydrology research;
 it is hardly available at hourly time step;

 it is often expensive for high resolutions;

Anyway, I consider that distributed hydrological models can take


full advantage of remote sensing observations that are able to well
describe the spatial distribution of variables. Thus, with the aim to
validate remote sensing data in future works, I designed a model
that allows their usage.
The adopted modular approach, in combination with the already
existent GRASS commands for remote sensing data, guarantees the
sufficient flexibility to use remote sensed data within the HydrFOSS
model.
Some of the existent GRASS command for remote sensing are:
 r.in.gdal, r.in.asci, r.in.bin, r.in.erdas, for importing;
 r.proj, i.group, i.target, i.points, i.rectify for geocoding;
 i.tm.dehaze, r.mapcalc, i.cluster, i.maxlik, i.class, i.gensig, i.smap,
for radiometric processing and classification.

4.1.2 Climatic map generation


In order to proceed with the estimation of the evapotransipration
and of the snowmelt, a number of distributed map (representing
climatic variables fields) has to be derived. Two steps have to be
done in order to proceed with the implementation:
1. generation of a meteorological database to archive
meteorological stations observations;
2. application of spatial interpolations.

4.2
Chapter 4 - System implementation

Meteorological network stations database (MNSD)


In the chosen approach the sampled data are point-wise
observations of climatic stations. The database schema that has
been adopted (following the logical model shown in paragraph
3.4.3) is represented by eight tables as represented in the relational
model of figure 4.1. It has to be noted that this database is not only
functional to the model development but also to the network
maintenance.

Figure 4.1 – Relational model of the point wise observations

In Appendix A.0 a detailed description of each table is presented.

Data interpolation
Climatological maps derivation results from the spatial interpolation
of measurements (temperature, humidity, solar radiation, wind
velocity, etc.) made at control points (meteorological stations).
Unfortunately these observation points are few and far from each
other while the investigated phenomenon has a high spatial
heterogeneity: as a result the correct field interpretation results
difficult.
For the precipitation and other meteorological variables
interpolation, the Thiessen’s polygons are still applied in many

4.3
Chapter 4 - System implementation

applications (e.g. Topkapi, WaSiM, Mike-SHE, and CASC2D):


Inverse Distance Squared Weighting (IDW) or Thiessen tessellation
(which is a particular IDW with number of interpolation points equal
to 1) are both available in GRASS through the s.surf.idw command.
The commonly adopted procedures, for meteorological variables
interpolation, have therefore been applied without any deeper
investigation except for the temperature (see chapter 5).
I think that, due to the high variability in space of the variables
(wind, rain, temperature, humidity, and solar radiation) and the
generally low spatial density of sampling, the improvements derived
by using more complex interpolation method (splines, kriging, co-
krigin, etc.) is often low compared to the increasing computational
cost.
The implemented procedure for meteorological raster maps
derivation consists in the access of the MNSD, in the extraction of
the desired informations and in the interpolation of the observations
by means of statistical, physical or mathematical modelling.
The database access and data extraction has been done by using the
db.connect, db.select, and v.in.ascii GRASS commands.
• db.connect enables the set up of the parameters needed by
GRASS for the connection to the databases;
• db.select allows to execute SQL query (red from from input file
or from standard input) printing the result to standard output;
• v.in.ascii imports a vector map in ASCII format (result of the SQL
selection) into current GRASS database.
For wind [m/s], rain [mm/h] and humidity [%] the Thiessen
tassellation is applied while for solar radiation [W·m-2·h-1] an IDW
interpolation is used, in order to avoid undesirable breakline, and
for temperature a linear gradient (chapter 5) is applied.
As an example the command's sequence to generate distributed rain
map for the 15th March 2005 at 13:00 is illustrated in frame 4.1.

4.4
Chapter 4 - System implementation

L1> #setting of connection with MNSD database


L2> db.connect driver=pg database=MNSD schema=""
group="";
L3> #extraction and generation of the vector of stations
and values
L4> echo “SELECT east, north, elev, value FROM
Measurement_points as p, Measurement as v WHERE
p.feature_id=v.feature_id AND v.time='15-03-2005 13:00'
AND v.param_id=3” | db.select | v.in.ascii
out=rain.200503151300 x=1 y=2 z=3 cat=0 format=point;
L5> #generation of the distributed map
L6> v.surf.idw -n input=rain.200503151300
output=rain.200503151300 npoints=1 col=dbl_1;
L7> #removing of the vector no more needed
L8> g.remove vect= rain.200503151300;
Frame 4.1 – procedure for rain map derivation.

Advantages of the GIS-embedded approach can be clearly detected,


in fact in order to change the interpolation method it will be enough
to change line 6 (L6) of frame 4.1 by using different GRASS
interpolation commands.

For example, the usage of spline interpolation can be easily


achieved by substituting L6 with the following line:

L6> v.surf.rst input=rain.200503151300 zcolumn=dbl_1;

As discussed in the previous chapter this procedure involves the


generation of a new vector file for each processing instant: even if
the file is later erased and therefore does not increase the
geodatabase size, it is certainly a computationally expensive
process.

4.5
Chapter 4 - System implementation

4.1.3 Watershed characterization maps generation


Meteorological distributed data (precipitation, temperature, wind
velocity, global radiation, etc.) are able to account for climatological
variability inside the catchment. Moreover land use, soil type, and
topography are required to completely parametrize the catchment.
In fact to model the hydrological processes of the catchment some
watershed characterization maps have to be derived: indexes
distribution mainly based on land use, soil properties and
topography.
For the radiation exchange between the earth surface and the
atmosphere, the albedo (a) of different surfaces and the Linke
Turbidy factor (Ltf) are required.
For the evapotranspiration process the status of the vegetation, in
terms of leaf area index (LAI), height (Vh), and density (Vc) must be
considered.
The solutions chosen for the creation of the maps of LAI, Vh and Vc,
according to the WaSiM model approach (Schulla, 1997), calculate
the raster map on the basis of values associated with four specific
days of the year and relative to a reference height of 400 m a.s.l.
(from [d1,400] to [d2,400] of the growing season and from [d3,400]
to [d4,400] of the dormant season) and relative to different land
uses.
Keeping in mind the existence of a temporal shift as a function of
the altitude, for each cell of the model with its elevation it is
possible to recalculate the reference days:

d i ,h = d i , 400 + 0.025 ⋅ ( h − 400 ) [eq. 4.1]

being i = [1,2,3,4] the index of the four reference days and h the
height above the sea level of the cell.
By means of the linear interpolation through time it is possible to
compute the value assumed by LAI, Vh and Vc cell by cell.
This algorithm has been implemented in the new GRASS command

4.6
Chapter 4 - System implementation

h.seasonal.par that – given a date, the four Julian days of reference


(j1, j2, j3, j4), the elevation model and the table that archives the
parameters reference values – computes all the desired maps.
The algorithm, implemented in a shell script language, reads the
input values from a seven columns table:
 column 1 – named PAR contains the parameter name;
 column 2 – named RASTER contains the reference raster map
name from which different values depends;
 column 3 – named CAT contains the category values of the
reference raster map;
 column 4 – named DAY1 contains the values of the parameters at
the reference day j1 for the current row PAR, RASTER and CAT
values;
 column 5 – named DAY2 is the same of column 4 but referred to
the day j2;
 column 6 – named DAY3 is the same of column 4 but referred to
the day j3;
 column 7 – named DAY4 is the same of column 4 but referred to
the day j4.
Then for each different parameter in the table (column PAR) it
calculates four raster maps by reclassifying the reference raster at
the four reference days, for each of them applies the equation 4.1 at
each cells, and finally, based on the input date, calculates the
desired parameter map with a linear interpolation through time.

On the other hand, a and Ltf do not undergo to any variation with
elevation and their mean monthly values are available in the
literature according to the different land uses. In this case the maps
do not depend on the elevation, but only on the date and on the land
use.
The derivation of these maps are computed by means of the new
developed GRASS command (Shell script language) h.montly.par
that reads the parameters values from a 15 columns table:

4.7
Chapter 4 - System implementation

 column 1 – named PAR contains the parameter name;


 column 2 – named RASTER contains the reference raster map
name from which different values depends;
 column 3 – named CAT contains the category values of the
reference raster map;
 column 4 to 15 – named JEN, FEB, MAR, APR, MAY, JUN, JUL,
AUG, SET, OCT, NOV, and DEC each one respectively contains
the values of the parameters for the twelve month relative to the
current row PAR, RASTER and CAT values;
Then for each different parameter in the table (column PAR), it
calculates the reclassified map based on the month extracted from
the input date.

4.2 Net radiation map generation


The net radiation (Rn) is the difference between the incoming net
shortwave radiation (Rns) emitted by the sun and the outgoing net
longwave radiation (Rnl) emitted by the earth.

4.2.1 Net shortwave radiation (Rns)


Following Hofierka and Suri (2002) the computation of the Rns has
been done in four steps:
 calculation of the net shortwave radiation components in clear-
sky condition. The existing r.sun command of GRASS computes
direct (beam), diffuse and reflected solar irradiation raster maps
for a given day, latitude, surface and atmospheric conditions.
Optionally a local time can be specified to compute solar
incidence angle and/or irradiance raster maps and the shadowing
effect of the topography can be incorporated (slope and aspect
input map can be derived from the r.slope.aspect command);
 Calculation of the total net shortwave radiation in clear-sky
condition (Rnsk). The sum of the direct (beam), diffuse and
reflected solar irradiation raster maps computes the total energy

4.8
Chapter 4 - System implementation

irradiation map;
 calculation of a clear-sky index (K) map. This is an index of
cloudiness that can be extracted from the interpolation of the
ratio between the estimated Rnsk and the radiation measured at
meteorological stations points; here the v.sample commands is
used to extract the Rnsk values, db.execute to calculate the K
index, and the v.surf.idw to interpolate K and obtaining the
distributed K field.
 calculation of the effective Rns by multiplying the Rns K in clear-
sky condition with the (1-K) factor by means of the map-algebra
command r.mapcalc.

The code to calculate the Rns map uses already existing GRASS
commands as shown in frame 4.2.

4.9
Chapter 4 - System implementation

L1> #calculation of Rsnk [W/m2]


L2> r.sun -s elevin=dem aspin=aspect slopein=slope lin=2
albedo=alb_Mar incidout=RADinc.200503151300
beam_rad=RADbeam.200503151300
diff_rad=RADdiff.200503151300
refl_rad=RADrefl.200503151300 day=73 time=13:00
dist=100;
L3> #GLOBAL Rsnk [W/m2]
L4> r.mapcalc
'Rnsk.200503151300=RADbeam.200503151300+RADdiff.200503151
300+RADrefl.200503151300';
L5> #setting of connection with MNSD database
L6> db.connect driver=pg database=MNSD schema=""
group="";
L7> #extraction and generation of the vector of stations
and values
L8> echo “SELECT east, north, elev, value FROM
Measurement_points as p, Measurement as v WHERE
p.feature_id=v.feature_id AND v.time='15-03-2005 13:00'
AND v.param_id=6” | db.select | v.in.ascii
out=RnskV.200503151300 x=1 y=2 z=3 cat=0 format=point;
L9> #sampling of the Rnsk (clear-sky condition)
L10> v.sample input=RnskV.200503151300 column=dbl_1
output=clear_sky.200503151300 rast=Rnk.200503151300;
L11> #removing no more used vector
L12> g.remove vect=RnskV.200503151300;
L13> #calculate the K index at sample points
L14> echo "UPDATE clear_sky.200503151300 SET diff = 0" |
db.execute;
L15> echo "UPDATE clear_sky.200503151300 SET diff =
(pnt_val/rast_val) where rast_val<>0" | db.execute;
L16> #interpolate K points and get K raster map
L17> v.surf.idw -n input=clear_sky.200503151300
output=clear_sky.200503151300 npoints=4 col=diff

4.10
Chapter 4 - System implementation

L18> #removing no more used vector


L19> g.remove vect=clear_sky.200503151300
L20> #calculation of Rns [W/m2]
L21> r.mapcalc 'Rns.200503151300 =Rnk.200503151300 *(1-
clear_sky.200503151300)'
Frame 4.2 – procedure for net solar radiation map calculation.

4.2.2 Net longwave radiation (Rnl)


Following Allen (1998), the longwave radiation can be computed in
relation to the surface temperature by the Stefan-Boltzman law
corrected with a humidity and a cloudiness factor:

Rnl = 2.043E-10 · [(T+273.16)4 · (0.34 - √ea) · (1.35 · (1-K) – 0.35)


[eq. 4.2]

with:
Rnl = net longwave radiation [MJ/(m-2∙h)];
T = surface temperature [°C];
ea = actual vapour pressure [KPa];
K = clear-sky index [-];

The actual vapour pressure can be computed as:

e0 = 0.618 · exp[(12.27·T)/(T+237.3)]
ea = (e0 · h)/100 [eq. 4.3]

where:
e0 = saturation vapour pressure [KPa];
ea = actual vapour pressure [KPa];
T = air temperature [°C];
h = relative humidity [%];

4.11
Chapter 4 - System implementation

The procedure to calculate the net longwave radiation makes


extensive use of the map algebra by means of the r.mapcalc
command as shown in frame 4.3.

L1> #saturation vapour pressure [KPa]


L2> r.mapcalc 'e0=0.6108*
exp((17.27*temp.200503151300)/(temp.200503151300
+237.3))';
L3> #actual vapour pressure[KPa]
L4> r.mapcalc 'ea=e0*humid.200503151300/100';
L5> #calculation of Rnl [MJ/(m2*h)]
L6> r.mapcalc 'Rnl.200503151300 =2.043e-
10*((temp.200503151300+273.16)^4)*(0.34-
0.14*sqrt(ea))*(1.35*(1-clear_sky.200503151300 )-0.35)';
L7>
Frame 4.3 – procedure for the long wave radiation map calculation

4.2.3 Global radiation (Rn)


The global net radiation, resulting from the incoming and outcoming
balance, can be computed by a simple map algebra operation (see
frame 4.4) where Rns is multiplied for 0.0036 in order to convert W
in MJ.

L1> #actual net radiation [MJ/(m2*h)]


L2> r.mapcalc 'Rn.200503151300 =(Rns.200503151300
*0.0036)-Rnl.200503151300';
Frame 4.4 – procedure for the global radiation computation

The entire procedure has been schematized in figure 4.2.

4.12
Datetime
DTM, slope, aspect
turbidity,,
Linke turbidity Clear
Clear--sky index
albedo

Rns
v.surf.idw
r.sun points v.sample Rns
r.mapcalc
obs..
obs

Solar irradiance

4.13
components r.mapcalc Rnsk r.mapcalc Rs
(bean diff.,
bean,, diff refl.)
., refl .)

r.mapcalc e0, ea r.mapcalc Rnl

Figure 4.2 – model of the Net Radiation map calculation.


Datetime
temperature, humidity
Chapter 4 - System implementation
Chapter 4 - System implementation

4.3 Snow accumulation and melt maps generation


The equations described in paragraph 3.3.3 have been implemented
in a new GRASS command: h.snow.
At first the algorithm estimates the liquid fraction (Fr) of the
precipitation at a given air temperature (T):

Fr = 1 if T>Tr;
Fr = (T+Tb )/(Tr +1) if Tb<T<Tr; [eq. 4.4]
Fr = 0 if T<Tb;

and calculates the solid (Ps) and liquid (Pr) parts:

Ps = R ∙ (1-Fr);
Pr = R ∙ Fr; [eq. 4.5]

whit R=rainfall [mm/h], Tr=3°C and Tb=-1°C.

Then it calculates the advected heat from precipitation (Qp) with


the equation:

Qp=0.001∙{[Ps∙Cs∙ρw∙min(T,0)]+
+[Pr∙(hf∙ρw+(CW∙ρw∙max(T,0)))]}; [eq. 4.6]

where ρw is the water density [1000 Kg∙m-³ ]; Cs the specific heat of


the ice [2.09 KJ∙Kg-1∙°C-1]; Ps the solid part of the precipitation
[mm/h]; Pr the liquid part of the precipitation [mm/h]; hf the heat of
fusion [333.5 KJ∙Kg-1]; and CW the specific heat of water [4.18 KJ∙Kg-
1
∙°C-1].

With this estimation (Qp) it is now possible to solve the equations


3.6 and estimates the new snowpack water equivalent Z*t+∆t and the
needed energy to maintain the snowpack in a solid phase E*t+∆t in a
zero snowmelt hypothesis.

4.14
Chapter 4 - System implementation

The energy needed to maintain the snowpack in a solid phase E n can


be calculated by means of:

En=Cs∙Z*(t+∆t)∙T0; [eq. 4.7]

where T0 is the ice temperature [273.5 K];


Now it is possible to evaluate if the available energy is enough to
melt part of the snowpack by comparing En with E*.
The algorithm calculates the snowmelt (Sm), the snowpack water
equivalent Zt+∆t , and the snowpack energy content Et+∆t for each cell
by setting:
 Sm=0, Zt+∆t=0 and Et+∆t=0 where there is no snow (Z*t+∆t=0);
 Sm=0, Zt+∆t=Z*t+∆t , and Et+∆t=E*t+∆t where En ≥ E*t+∆t (the available
energy is not sufficient to melt part of the snowpack);
 Sm=min((E*t+∆t-En )/Cf,Z*t+∆t), Zt+∆t=max(0,Z*t+∆t-Sm), and
Et+∆t=E*t+∆t-((Cs∙T0+hf)∙Sm) where En < E*t+∆t (the available energy
is sufficient to melt part of the snowpack);

The new maps of Sm, Zt+∆t , and Et+∆t are finally generated. The
input-output schema of the h.snow command is shown in figure 4.3.

Rn Snowpack water equivalent


Temperature h.snow Snowpack energy content
Rainfall Snowmelt
Snopack water eqivalent

Figure 4.3 – input-output schema of the h.snow command.

4.15
Chapter 4 - System implementation

4.4 Potential evapotranspiration map generation


Following the approach described in paragraph 3.3.1, I developed
the new GRASS command h.evapo.PM that, given the vegetation
height (hc), humidity (Rh), wind speed at two meters height (u2),
temperature (T), digital terrain model (Z), and net radiation (Rn)
raster input maps, calculates the potential evapotranspiration map.
The implemented algorithm applies the open water calculation
(equation 3.2) where the vegetation height is negative and the
Penman-Monteith (equation 3.1) elsewhere.
The equation 3.1 and 3.2 can be solved after the computation of the
single variables found in those formulas:

 mean saturation vapour pressure ea [KPa]:


ea = 0.61078∙exp[(17.27∙T)/(T+237.3)]; [eq. 4.8]

 slope of vapour pressure curve ∆ [Kpa/°C]:


∆ = (4098∙ea)/(237.3+T)2; [eq. 4.9]

 latent heat vaporisation λ [MJ/kg]:


λ = 2.501 – (0.002361∙T); [eq. 4.10]

 atmospheric pressure P [KPa]:


g/(η∙R)
P = P0 ∙[(Tk0-η∙(Z-Z0))/Tk0] ; [eq. 4.11]

 psichiometric constant γ [KPa/°C]:


γ = 0.001∙[(cp∙P)/(ε∙λ)]; [eq. 4.12]

 aerodynamic resistance ra [s/m]:


where the vegetation height (hc) is less than 2m:
d= (2/3)∙hc; Zom=0.123∙hc; Zoh=0.1∙Zom;
ra=[log((Zw-d)/Zom)∙log((Zh-d)/Zoh)]/ (k2∙u2); [eq. 4.13]

4.16
Chapter 4 - System implementation

elsewhere:
u10 = u2∙(log((67.8∙10)-5.42))/4.87;
ra = 94 / u10; [eq. 4.14]

 surface resistance rs [s/m]:


rs = 100/(0.5∙24∙hc); [eq. 4.15]

 modified psichiometric constant γ ∗ [KPa/°C]:


γ ∗ = γ ∙(1+(rs/ra)); [eq. 4.16]

 soil heat flux G [MJ/m2*d]:

during the day: G=0.1∙Rn; [eq. 4.17]

during the night: G=0.5∙Rn; [eq. 4.18]

 actual vapoure pressure ed [KPa]:

ed = Rh∙ea/100; [eq. 4.19]

 virtual temperature Tkv [°C]:

Tkv = (T+273.15)/(1-(0.378∙ed/P)); [eq. 4.20]

 atmospheric density ρ [Kg/m3]:


ρ = P/(Tkv∙R/100); [eq. 4.21]

where:
cp is the specific heat of moist air [1.013 KJ∙Kg-1∙°C-1];

ε is the ratio of molecular weigth of water to dry air [0.622];


P0 is the atmospheric pressure at sea level [101.3 KPa];
Tk0 is the reference temperature at sea level [293.16 K];

η is the constant lapse rate [0.0065 K/m];


Z0 is the altitude at sea level [0 m];
Z is the elevation above Z0;

4.17
Chapter 4 - System implementation

g is the gravitational acceleration [9.81 m/s];


R is the specific gas constant [287 J∙kg-1∙K-1];
Zw is the height of wind measurements [2 m];
Zh is the height of humidity measurements [2 m];
k is the Von Karman constant [0.41];
d is the zero plane displacement of wind profile [m];
Uz is the windspeed at height z [m/s];
Zom is the roughness parameter for momentum [m];
Zoh is the roughness parameter for heat and water vapour [m];

Optionally the user can activate a flag (-z) that allows him setting to
zero all of the negative evapotranspiration cells; in fact these
negative values motivated by the condensation of the air water
vapour content, are sometime undesired because they can produce
computational problems.
The usage of the flag -d detect that the module is run in day hours
and the appropriate soil heat flux is calculated.
The input-output schema of the h.evapo.PM command is shown in
figure 4.4.

LAI
Vegetation height
Vegetation density Potential
Humidity h.evapo.PM evapotranspiration
Wind speed
Net Radiation

Figure 4.4 – input output schema of the h.evapo.PM command.

4.18
Chapter 4 - System implementation

4.5 Canopy interception maps generation


The canopy interception mechanism is described in the paragraph
3.3.2. In order to proceed with the interception losses estimation,
the potential evaporation from a liquid surface (ETPls) has to be
known. Thus the previously described command h.evapo.PM has to
be run with a “virtual” map of the vegetation height set to negative
values. Given the LAI [-], the ETP ls [mm/h], the leaf storage level at
the previous step C0 [mm], the rain R [mm/h], the timestep Ts [h],
and the vegetation density Vc [%] the command estimates the
canopy storage maximum capacity Cm [mm]:

Cm = 0.35∙LAI [eq. 4.22]

and the needed time to evaporate a saturated canopy τ [h]:

τ = Cm /ETPls; [eq. 4.23]

if the ETPls is zero, in order to avoid division by zero it is simply set


to a large value:

τ = Ts∙1000 [eq. 4.24]

As Zengh (2000) discussed, the storage level dynamic during a


rainfall event can be splitted in three different phases (see figure
4.5):
A) filling – it is raining and the canopy storage is not saturated, thus
the storage level is growing;
B) saturation – it is raining and the canopy storage has reach the
maximum capacity, thus its level remain at the maximum;
C) drying – it is not raining and the canopy storage is not empty,
thus the level is going down.

4.19
Chapter 4 - System implementation

Ca

B
A
C
TS
Time

Figure 4.5 – canopy storage level dynamic during storm: filling (A),
saturation (B), and drying (C).

Depending on the storage level and rain duration/intensity during a


storm all the three phases (A+B+C, if saturation is reached) or only
two of them (A+C, if saturation is not reached) can occur.
As a function of the phases taking place into the single time step (t 0–
t), different solutions of the equations 3.3, 3.4, and 3.5 has to be
applied.
Thus the algorithm at first, being C0 the storage level at t0,
computes the rain needed to saturate the storage at the end of the
interval (Rsat ):

Rsat = (C0-Cm)/[τ∙(1-exp(-Ts/τ ))] [eq. 4.25]

and the time when saturation is reached (dt):

dt = -τ∙log[1-((Cm-C0)/(R∙τ))] [eq. 4.26]

Then it compares R [mm/h], ETPls [mm/h], C0 [mm], Cm [mm], Rsat


[mm/h], and dt [h] evaluating 5 different cases.
For each occurring case the algorithm estimates the interception
losses IL [mm], the drainage D [mm] and the canopy storage level at
the end of the step Ct [mm] as follow:

4.20
Chapter 4 - System implementation

 Case 1 – only saturation phase (R>ETPls and C0≥Cm ):


IL = min[(ETPls∙Ts),(R∙Ts)+(C0-Cm)]
D = R∙Ts - IL + (C0-Cm)
Ct = Cc [eq. 4.27]

 Case 2 – only filling phase (R>ETPls and C0≤Cm and Rsat>R):


Ct = R∙τ∙(1-exp(-Ts/τ )) + d_W0;
IL = (R∙Ts) - (Ct-C0);
D = 0; [eq. 4.28]

 Case 3 – mixed filling and saturation phase (R>ETP ls and C0≤Cm


and Rsat≤R):
IL = (R∙Ts) - (Cm-C0)+(Ts-dt)∙ETPls
D= (R∙Ts) - (Cm-C0) - IL;
Ct = Cm [eq. 4.29]

 Case 4 – only drying phase (0<R<ETPls) or (R=0 and C0≤Cm):


D = 0;
Ct = max(W0∙exp(-Ts/τ ),0);
IL = (C0-Ct)+R; [eq. 4.30]

 Case 5 – empty phase (R=0 and C0 =0):


IL = 0;
D = 0;
Ct = 0; [eq. 4.31]

The input output schema of the implemented command


h.interception is shown in figure 4.6.

4.21
Chapter 4 - System implementation

Potential evapotraspiration
from liquid surface Interception losses
Time step h.interception Drained rain
Rain Final canopy storage leve
Initial canopy storage level

Figure 4.6 – input output schema of the h.evapo.PM command.

4.6 Rainfall runoff maps generation


The governing equation for the runoff module has been presented in
paragraph 3.3.4.
Solving those formulas implies the knowledge of the values of the
following parameters at each cell:
 saturated hydraulic conductivity ks [m/s] – soiltype dependent
factor;
 saturated soil water content θs [m3/m3] – soiltype dependent
factor;
 residual soil water content θr [m3/m3] – soiltype dependent factor;
 soil type dependent elevation parameter α [-];
 soil thickness of the soil affected by subsurface flow L [m] –
soiltype dependent factor;
 manning friction coefficient n [m-1/3s-1] – landuse dependent
factor;
 terrain slope β [deg];
 terrain hydrological type ht [-] – terrain or river;
 river width [m];

The terrain hydrological type allows to select which equations has


to be solved, with this aim a map has to be derived. The river cells
(where subsurface and eventually overland flow equation will be
solved) will be identified by the value 1, while the terrain cells
(where river, subsurface and eventually overland flow equation will
be solved) will be identified by the value 0. Further development of

4.22
Chapter 4 - System implementation

the runoff module, which may include lake or artificial basin


modelling, could be achieved by using different hydrological type
values.

Furthermore additional information are needed to solve in the


correct order the flow that drains from one cell into another:
 flow direction – that shows in which of the 8 neighbour cell the
water drains;
 flow accumulation – that shows the number of up-cells draining
into.

Finally, with the aim to prevent any out of memory operations, each
basin has to be solved separately but in a consequential order that
permits to consider the outflow of a basin as one inflow for the
following; thus the last information needed is:
 basin number – unique basin identifier.

Given a digital terrain model (DTM), the slopes, flow directions, flow
accumulations, rivers and basins maps can be derived with the
r.solpe.aspect and the r.watershed GRASS commands.
In order to correctly extract these maps with the r.watershed
command a correct threshold value has to be chosen, in fact this
value affects both the river and basin delineation. The right value
can be chosen by visually comparing different r.watershed output
river maps (obtained with different threshold values) with an
existing vector map and selecting the one that gives the best fitting.
Even if r.watershed do not ask for an elevation map without sinks,
the result can be slightly different in the two cases (original DTM or
filled DTM as input). The user can check if a filled DTM can improve
the results by generating a filled elevation map with the GRASS
r.fill.dir or r.terraflow command.
Note that for sake of simplicity the runoff are solved only in the four
main directions, thus the r.watershed command has to be run with

4.23
Chapter 4 - System implementation

the flag -4 that allows only horizontal and vertical flow of water.
Future studies could implement other more elegant solution with
multiple flow direction.

4.6.1 Initialization command: h.hydroFOSS.init


The flows (subsurface, surface and channel) have to be computed
cell by cell following the flow accumulation order: this ensures that
every incoming discharge has been already estimated, but also
implies that raster cells have to be accessed in a sparse order.
These tasks are commonly performed by using temporary files or by
loading the entire map into the memory. While the first method is
time consuming because it needs multiple file position indicator
settings (by means of the C ANSI fseek function), the second could
imply limitation in solving large watersheds due to insufficient
physical memory size.
In order to skip these two drawbacks the chosen approach makes
extensive use of the database: this allows extracting the basins, and
within each basin the cells with their own associated values, in the
correct order of solution.
To benefit of the database properties a table with the previously
mentioned informations has to be generated: the new developed
command h.hydroFOSS.init has been designed for this purpose.
The input raster maps containing the cells properties and values for
the cells of the region are: the filled DEM (without sinks), the
slopes, the basins, the flow directions, the soiltypes, the landuses,
the hydrotypes, the flow accumulations, and the riverwidths.
Furthermore the algorithm extracts informations from two tables,
named soiltype_cat and landuse_cat, where soiltype and landuse
dependent parameters values are stored.
The soiltype_cat table contains six columns named cat (the soiltype
map category values), kappa_s (the saturated hydraulic conductivity
values), teta_s (saturated soil water content values), teta_r (residual
soil water content values), alfa (soil type dependent values), thik

4.24
Chapter 4 - System implementation

(the thickness of the soil affected by subsurface flow values).


The landuse_cat table contains two columns, named lu_id (the
landuse map category values) and manning (the Manning friction
coefficient). The h.hydroFOSS.init uses the values found in the input
raster maps and in the two tables to populate the initialization table
that will be accessed for discharge maps computation.
The overall process that brings to the generation of the initialization
table is represented in figure 4.7.

Filled DEM
r.fill.dir
soiltype_cat table
landuse_cat table
landuse
r.watershed soiltype

Basins
Flow accumulations
DEM h.hydroFOSS.init Initialization table
Flow directions
Rivers

Figure 4.7 – rainfall runoff initialization procedures.

The output initialization table will store the informations of one cell
for each row and will have the following attributes:
 cat [int4] – the cell index Cidx. Referring to figure 4.8, the cell
index is a progressive integer number calculated with reference
to the current GRASS region by the formula:
Cidx=(ncolumns*i)+j [eq. 4.32]
where ncolumns are the total columns of the region, i is the
GRASS row index and j is the GRASS column index;
 ks [float8] – the saturated hydraulic conductivity ks;
 teta_s [float8] – the saturated soil water content θs;
 teta_r [float8] – the residual soil water content θr;
 alfa [float8] – the soil type dependent α;
 L [float8] – the soil thickness L;
 n [float8] – the manning coefficient n;

4.25
Chapter 4 - System implementation

 basin [float8] – the basin index Bidx: a unique index of the basin
whose the cell belongs;
 direction [float8] – the flow direction index F idx: provides the
drainage direction values as defined in the r.watershed
command.
 accumulation [float8] – the flow accumulation value A cc: number
of up-cells that drain into each cell as defined in the r.watershed
command.
 slope [float8] – the terrain slope β: value, stated in degrees of
inclination from the horizontal, as described in the r.slope.aspect
command;
 h_type [float8] – the hydrological type value: currently only two
values are defined “0” for terrain and “1” for river;
 width [float8] – the riverwidth value;
 l_use [int4] – the landuse category value lu cat: the coded class
which the cell belongs;
 s_type [int4] – the soiltype category values stcat: the coded class
which the cell belongs;

West region limit East region limit

North
0 1 2 3 4 5 6 7 region limit
region
resolution 8 9 10 11 12 13 14 15
16 17 …
ith row cat

South N nrows
region limit
jth column ncolumns
Figure 4.8 – cells indexing in cat attribute of the initialization table
outputted from the h.hydroFOSS.init command.

4.26
Chapter 4 - System implementation

4.6.2 Runoff computation command: h.hydroFOSS.runoff


The runoff computation algorithm has been developed into the new
module h.hydroFOSS.runoff.
Before to proceed with the runoff estimation, the correct amount of
water available for the routing (AW) and the actual
evapotranspiration (ETa) rate has to be calculated.
The AW can be calculated as the water drained from canopy
interception plus the rainfall directly falling over the ground plus
the snowmelt.
The ETa can be estimated following a Topmodel approach. It applies
a simple reduction function to the potential evapotranspiration
(ETp) on the base of the soil water content as follows:

ETa = (ETP·Vs)/(ξ·Vsat) if Vs≤ ξ·Vssat


ETa = ETP if Vs> ξ·Vssat [eq. 4.33]

where Vs is the soil water content, Vssat is the maximum soil water
content and ξ is a threshold.
Following Menzel (1997) the value for the threshold ξ can be fixed
equal to 0.6, that means to estimate evapotranspiration at potential
rate for soil saturated with more then the 60% of their capacity and
reduced otherwise.
Both, the AW and the ETa can be estimated by applying the map
algebra as shown in frame 4.5.

4.27
Chapter 4 - System implementation

L1> #runoff water estimation [mm/h]


L2> r.mapcalc 'NETrain.200503151300=Idrain.200503151300 +
snowmelt.200503151300 +
(1-vegc_73)*rain.200503151300;
L3> #actual evapotranspiration estimation [mm/h]
L4> r.mapcalc 'EREAL.200503151300 =
if((soil_wc.200503151300
<(0.6*soil_wc_max)),(EPOT.200503151300*soil_wc.2005031513
00/0.6*soil_wc_max)),EPOT.200503151300 )';
Frame 4.5 – map algebra application for runoff water and actual
evapotranspiration estimation.

Once calculated:
 the initialization table (generated with the h.hydroFOSS.init
command);
 the AW and ETa map;
 the starting overland water content (Vo);
 the starting soil water content (Vs);
 the starting channel water content (Vc);
the new developed GRASS command h.hydroFOSS.runoff can be
run.
The h.hydroFOSS.runoff algorithm flux can be described by the
following steps:
1. selection of each basin index ordered by the maximum
accumulation, the minimum and maximum row, the minimum and
maximum column, and the minimum and maximum accumulation
values.
2. allocation of a vector for storing the parameters of the outlet of
each basin;
3. starting of a loop trough the ordered basins;
4. allocation for the current basin of a matrix ranging from
minimum column-1 to maximum column+1 and from minimum
row-1 to maximum row+1. A frame of one cell around the exact
matrix containing the basin guarantees the inclusion of every

4.28
Chapter 4 - System implementation

inlet (outlet of previously processed basin);


5. accordingly to the flow accumulation order, selection of all the
cells, with their respective properties, that belong to the
currently processed basin;
6. population of the allocated matrix with the outlet falling into the
matrix itself;
7. starting of a loop through the selected cells in flow accumulation
order;
8. solution for the currently processed cell of the equations 3.10,
3.11, and 3.12 with a Runge-Kutta procedure;
9. storing of the results for the currently processed cell in the
allocated matrix;
10.once solved the flow equations for all the cell of the currently
processed basin, writing of the estimated values stored in the
allocated matrix into temporary files and go on throw the basin
loop by processing the following basin (go back to step 4).
11.once solved all the basins, and transformed all the temporary
files in permanent files the (i) soil water content [m3], (ii)
overland water content [m3], (iii) channel water content maps
[m3], and optionally (iv) soil water discharges [m3/s], (v) overland
water discharges [m3/s], and (vi) channel water discharges [m3/s]
maps, calculated at the end of the timestep, are outputted.

As shown in figure 4.9, the solution of the runoff equations for the
cells where the river does not cover all the cell area, implies the
solution of the overland and the subsurface equation over an area
equal to the cell resolution minus the riverwidth, and the solution of
the channel equation for an area equal to the riverwidth.
A further problem in these cells is represented by the Manning
coefficient definition: in fact the initialization table assigns one
value for each cell while two values are needed (one for the channel
and one for the soil) in order to solve the flow equations.
In my implementation the archived Manning values for those cells

4.29
Chapter 4 - System implementation

crossed by river are referred to the channel, and the values referred
to the soil part of the cell is estimated as the mean values of the not
river neighbours cells.

Figure 4.9 – h.hydroFOSS.runoff solved water flux over two cells.

The kinematic differential equations of flow have to be solved in the


following order:
1. subsurface: in fact all the AW instantaneously goes into this layer
until it is saturated; solving this equation it is possible to estimate
the soil water discharges and the eventual ex-filtration to the
overland layer.
Being:
VSi = the soil water content of the ith cell;
QSi = the soil water discharge of the ith cell;
pi = the runoff water of the ith cell;
X = the grid resolution;
QOiU = the overland discharges of the cells drained by the ith cell;

4.30
Chapter 4 - System implementation

QSiU = the subsurface discharges of the cells drained by the ith


cell;
ETai = the actual evapotranspiration for the ith cell;
CSi = the subsurface local conductivity coefficient for the i th cell
as defined in equation 3.9;
rOi = soil water excess for the ith cell;
the subsurface equation results:

∂VSi C ⋅X [eq. 4.34]


∂t
(
= pi X 2 + QOi
U

X
)
+ QSiU − Si 2α S ⋅VSiα S

Solving the equation 4.33 and subtracting the ETai the Vsi(t+∆t),
the QSi(t+∆t) and the rOi(t+∆t) can be calculated.

2. Overland: only if the ex-filtration process is active the overland


flow occurs and the overland equation can be solved in order to
estimate the overland discharges;
Being:
VOi = the overland water content for the ith cell;
QOi = the overland water discharges for the ith cell;
COi = the coefficient relevant to the Manning formula for the ith
cell;
the overland equation results:

∂VOi C ⋅X
∂t
( X
)
= rOi X 2 − Oi2α O ⋅ VOiα O [eq. 4.35]

Solving equation 4.35, the VOi(t+∆t) and the QOi(t+∆t) can be


estimated.

3. Channel: if the cell contain the river, the two subsurface and
overland flows are used as input for channel discharge
estimation.
Being:

4.31
Chapter 4 - System implementation

VCi = the channel water content for the ith cell;


QCi = the channel water discharges for the ith cell;
CCi = the coefficient relevant to the Manning formula for the ith
cell;
Wi = the river width for the ith cell;
QCiU = the channel discharges of the cells drained by the ith cell;
can be obtained the equation:

∂VCi
( ) C ⋅W
= rCi ⋅ X ⋅ Wi + QCiU + QSi + QOi − Ci αi C ⋅ VCiα C [eq. 4.36]
∂t ( X ⋅Wi )

Solving equation 4.36, the VCi(t+∆t) and the QCi(t+∆t) can be


estimated.

Figure 4.10 shows the input-otput schema for the


h.hydroFOSS.runoff command.

initialization table overland water volume


runoff water soil water volume
actual evapotraspiration channel water volume
starting overland water volume h.hydroFOSS.runoff overland water discharge
starting soil water volume soil water discharge
starting channel water volume channel water discharge
timestep saturation water excess

Figure 4.10 – input output schema of the h.hydroFOSS.runoff command.

It has to be noted that in order to select the optimal basin size to be


processed at one time, the obtained basins map can eventually be
aggregated in larger basins that join up-flow basins. This operation
can be done by checking the maximum flow accumulation of
neighbours basins and aggregating those with lower values.

4.32
Chapter 4 - System implementation

4.7 Automatic model running


The procedures and the commands described in the previous
paragraphs have to be run iteratively for each time step, that for the
current model has been chosen of one hour.
This task is achieves by implementing the Shell script
h.hydroFOSS.model that loops, from a starting date-time to an
ending date-time, through the time and performs all the needed
operations (catchment status maps generation, net radiation map
generation, snow accumulation and melt maps generation, potential
evapotranspiration map generation, canopy interception maps
generation, and rainfall runoff maps generation).
The awk print function allows to parse the input dates (defining the
starting and ending simulation period) and to set up the date suffix
used to named the raster series maps.

>DATEstrFROM=`echo $GIS_OPT_from | awk -F"-" '{print($1" "$2"


"$3" "$4" "$5" 00");}'`
>DATEstrTO=`echo $GIS_OPT_to | awk -F"-" '{print($1" "$2" "$3"
"$4" "$5" 00");}')`

Then the while loop ending when through hours is achieved by


using the shell function mktime, strftime as follow:

>while [ "$DATEstrFROM" != "$DATEstrTO" ] ; then


>DATE_NOW=`gawk "BEGIN{print
strftime(\"%Y%m%d%H%M\",mktime(\"$DATEstrFROM\")) }"`
>DATE_LAST=`gawk "BEGIN{print
strftime(\"%Y%m%d%H%M\",mktime(\"$DATEstrFROM\")-3600) }"`
>..........
>DATEstrFROM=`gawk "BEGIN{print strftime(\"%Y %m %d %H
%M 00\",mktime(\"$DATEstrFROM\")+3600) }"`
>fi

4.33
Chapter 4 - System implementation

The number of generated maps can highly increase during the


model running through time (i.e. 16376 a mount), causing two
problems:
• for Linux distributions using a kernel version younger than the
2.4, large numbers of files and folders in a single directory can
lead to system errors due to the existence of a maximum node
limit (inode-max: describes the maximum number of files or
folders allowed for each directory);
• large numbers of nodes cause loss of performance in accessing
files.
To skip these problems the directory structure schema shown in
figure 4.11 has been adopted.
The active location has a “root mapset” containing all the basic
maps like DTM, landuse, soiltype, riverwidth, etc. and a saved
region that defines the model domain (root_mapset of figure 4.11).
Then the model creates for each month a new mapset (identified by
the root mapset name plus a month-year suffix: i.e.
mapsetX.aug2005) that will store all the maps of the raster series
belonging to that month.

4.34
Chapter 4 - System implementation

grassDB

location1

PERMANENT

root_mapset

root_mapset.aug2005

root_mapset.sep2005

root_mapset.oct2005

rain.200510010000
rain.200510010100
Grass database
rain.200510010200
Grass location .
Grass mapsets .
Grass maps .
rain.200510312300
Figure 4.11 – hydrofoss directory structure.

For each modelled time step, h.hydroFOSS.model calculates the


current date-time (DTc) and the previous step date-time (DTp), the
current mapset (Mc) and the previous step mapset (Mp).
It is now possible to:
• access the last calculated maps by adding as a suffix the DTp to
raster series name, and the Mp as the mapset where they reside
(i.e. with raster series name = “rain”, Dtp=“200510171300”, and
Mp=“oct2005”: map to read is “rain. 200510171300@ oct2005”);
• select the database informations by using the DTc
(i.e. with Dtc=200510171400:
“SELECT east, north, elev, value
FROM Measurement_points as p, Measurement as v
WHERE p.feature_id=v.feature_id

4.35
Chapter 4 - System implementation

AND to_char(v.time,'YYYYMMDDHH24MI')='200510171400'
AND v.param_id=3”);
• generate new rasters maps and add to the raster series, in the
Mc mapset, by using the DTc suffix
(i.e. with raster series name = “rain”, Dtc=“200510171400”, and
Mc=”oct2005”: mapset to be set as active (g.mapset) is
”oct2005” and map to be created is “rain. 200510171300”);

The inputs required to run h.hydroFOSS.model are:


 general:
root mapset name, and model region name;
 rasters:
filled DEM map name, landuse map name, soiltype map name,
aspect map name, and slope map name;
 database:
climatic station database name, database driver name, table
name for seasonal parameters, table name for monthly
parameters, a string comma separated values of the four
reference days, and initialization table name.

The generated time series maps are:


 rain; wind; humidity; temperature; directed, diffuse, reflected,
and global solar radiation; clear sky index; net solar radiation;
saturation vapour pressure; actual vapour pressure; longwave
radiation; global radiation; snowpack water equivalent; snowpack
energy content; snowmelt; potential evapotranspiration; potential
evapotranspiration from a liquid surface; canopy storage level;
interception loss; canopy drainage; water available for runoff;
real evapotranspiration; overland discharge; subsurface
discharge; channel discharge; overland water content;
subsurface water content; and channel water content.

4.36
Chapter 4 - System implementation

4.8 Sensitivity analysis and model calibration


Even if HydroFOSS is a physical model, the values used to
characterize the watershed are mean values calculated over the
cells and are not directly measurable.
Therefore a sensitivity analysis and a calibration of the runoff
component is needed.

4.8.1 Advanced calibration method


UCODE-2005 (Poeter et al., 2005) was used to compute
sensitivities, parameter correlation coefficients, and to calibrate the
hydrological model. The sensitivities are calculated by UCODE-2005
using parameter perturbation methods. UCODE-2005 is developed
using the JUPITER API capabilities and conventions, which
facilitates inter program communications.

In recent years, automatic calibration is becoming more diffused


also for the calibration of hydrological model (Madsen, 2000). In
manual calibration, a trial-and error parameter adjustment is made.
In this case, the goodness-of-fit of the calibrated model is basically
based on a visual judgement by comparing the simulated and the
observed hydrographs. Accordingly to Madsen (2000), “For an
experienced hydrologist it is possible to obtain a very good and
hydrologically sound model using manual calibration. However,
since there is no generally accepted objective measure of
comparison, and because of the subjective judgement involved, it is
difficult to assess explicitly the confidence of the model simulations.
Furthermore, manual calibration may be a very time consuming
task, especially for an inexperienced hydrologist.”

Using the automatic calibration, parameters are adjusted


automatically, according to specified rules and numerical measures
of the goodness of the fit.

4.37
Chapter 4 - System implementation

UCODE-2005 can be applied with any application model or set of


models; the only requirement is that they have numerical (ASCII or
text only) input and output files and that the numbers in these files
have sufficient significant digits. The estimated parameters can be
defined with user-specified functions: for example prior, or direct,
information on estimated parameters also can be included in the
regression. Observations to be matched in the regression can be
any quantity for which a simulated equivalent value can be
produced, thus simulated equivalent values are calculated using
values that appear in the application model output files and can be
manipulated with additive and multiplicative functions, if necessary.
The nonlinear regression problem is solved by minimizing a
weighted least-squares objective function with respect to the
parameter values using a modified Gauss-Newton method.
Sensitivities needed for the method are calculated approximately by
forward or central differences and problems and solutions related to
this approximation are discussed. Statistics are calculated and
printed for use in (1) diagnosing inadequate data or identifying
parameters that probably cannot be estimated with the available
data, (2) evaluating estimated parameter values, (3) evaluating the
model representation of the actual processes and (4) quantifying the
uncertainty of model simulated values.

4.38
Chapter 4 - System implementation

4.8.2 Calibration procedure implementation


The runoff model uses parameters stored in a Database Table.
For each cell, the table (see table 4.1) stores the values of the
needed parameters.

Column Type Description


cat integer cell_ID
ks double precision conductivity
teta_s double precision saturated soil moisture content
teta_r double precision residual soil moisture content
alfa double precision soil parameter dependent
l double precision soil thickness
n double precision manning coefficient
basin integer basin_ID
direction integer flow direction
accumulation integer flow accumulation
slope double precision slope
h_type integer hydrological cell type (river/soil)
width double precision river width
landuse integer landuse map category
soiltype integer soiltype map category
Table 4.1 – initialization table fields

The model generates for each hour a new map of discharges.

The UCODE program is run with the configuration file (.in) as an


argument.
The .in file specifies:
 the command to run the model;
 the parameters to be evaluated and their information (starting
value, lower reasonable value, largest reasonable value, scale
parameter values, adjustability (yes/no), fractional amount of
perturbation for sensitivity, applied log-transformation (yes/no),
maximum fractional change between iterations, etc.;

4.39
Chapter 4 - System implementation

 the observation data file (.flo);


 the model input file (.if);
 the model input file template (.tpl);
 the model output file (.est);
 the model output file structure (.ins).

In order to link the model with the calibration program the


following task are required:
1. create a command that samples the channel discharge raster in
the given points and creates the estimation file (.est).
The command h.sample.pt has been generated by modifying the
already existing GRASS command r.sample: the new command
given a raster and a point vector map, prints to standard output,
for each point: the point cat value, the raster name, and the value
of the cell where the point falls into.
i.e. “h.sample.pt vector=st raster=channel_wd.200408022200”
prints out:
(1)channel_wd.200408022200 | 1.306722
(2)channel_wd.200408022200 | 5.317407
2. Create the output file structure files (.ins), to allow UCODE to
extract estimations. This is done following the UCODE
requirements for each observation by setting the line number,
the observation name, and the character of the line containing
the value (starting character: ending character)
i.e.
jif @
1 [(1)channel_wd.200408022000]18:22
2 [(1)channel_wd.200408022100]18:22
3 [(1)channel_wd.200408102200]18:22
.....
3. create a template file for update the parameters.
This is the file that the model runs to change the parameter
values at each new call; in practice being the parameters in a

4.40
Chapter 4 - System implementation

database table, this is an SQL file with the parameters name


instead of the values.
i.e.
UPDATE IniTab SET ks = @ ks_soiltype_1 @ WHERE soiltype=1;
UPDATE IniTab SET ks = @ ks_soiltype_2 @ WHERE soiltype=2;
UPDATE IniTab SET l = @ l_soiltype_4 @ WHERE soiltype=4;
UPDATE IniTab SET n = @ n_landuse_11 @ WHERE landuse=11;
.......
4. create the observation files to be used to compare the model
estimations and perform the calibration (.flo). This is the files
where the river gauges observations are written.
i.e.
BEGIN OBSERVATION_DATA TABLE
NROW=1440 NCOL=6 COLUMNLABELS
(1)channel_wd.200408102100 0.78537
(1)channel_wd.200408102200 0.81421
(1)channel_wd.200408102300 1.13655
....
5. create a command that updates the parameters and run the
model. This is a simple shell script, calling the db.execute GRASS
command, to executes the model input file (.if) generated by
UCODE and then the h.hydroFOSS.model.
i.e.
> db.execute input=filename.if;
> h.hydroFOSS.model date_from=200408011200
date_to=200408311200 db=verzasca drv=pg tab=ucode_test
dem=dem_fill slope=SLOPE soiltype=soiltype landuse=landuse
mapset=verz_cal region=verzasca sample=verz_lim@verz_cal
seas_tab=season_par seas_days=”120,180,230,290”
mon_tab=monthly_par;

It is now possible, in a Shell windows, to run within GRASS the


calibration process by calling the command:

4.41
Chapter 4 - System implementation

> ucode config.in calibration1;

The described procedure realizes the linkage between GRASS and


UCODE-2005, its application will be shown in the chapter 6, where
a first calibration has been run.

4.42
Chapter 5 - Investigation of the radar rainfall estimates
and of the temperature interpolation method

5. Investigation of the radar


rainfall estimates and of the
temperature interpolation method
As previously discussed in the section 4.1.1 in this chapter a
detailed investigation on the estimations of rainfall intensity derived
from meteorological radar observations (paragraph 5.1) and the
temperature interpolation technique (paragraph 5.2) are presented.

5.1 Radar rainfall estimates investigation


Remote sensed data are attractive because they can be sampled in
a distribute way but often, unlike the gauges observations, are not
direct measurements of the phenomena.
In order to validate the derived information, a comprehensive
analysis of rainfall estimations, by means of weather data in
Southern Switzerland and Northern Italy, is carried out. Validation
of the MeteoSwiss weather radar data vs. rainfall gauges data is
conducted by using a data set composed of 365 days (Year 2002)
per 24 hourly observations at 35 different rain gauges: these have
been homogenized by classifying gauge measurements on the basis
of the radar estimates classes and further subdivided in two
different seasonal datasets.

5.1.1 State of the art in rainfall estimate


For operational real-time flood forecasting systems ground-based
rain-gauges are still commonly used for several reasons: (i) they
provide a direct measurement of rain, although in a single point; (ii)
the measurement and the data transfer techniques are nowadays
well established; (iii) radar and satellite systems still require high
investment and maintenance cost. In the last 20 years increasing
attention has been focused on weather radar system for spatial

5.1
Chapter 5 - Investigation of the radar rainfall estimates
and of the temperature interpolation method

measurement of precipitation. In the 1980s the EU sponsored the


COST72 (1985) and COST73 programs, with the aim of establishing
a European radar network.
The COST Action 717 involves radar experts, hydrologists, and
meteorologists from several countries dealing with “Use of Radar
Observations in Hydrological and Numerical Weather Prediction
(NWP) models”. A common basis of all activities within the project
is the availability of reliable radar-derived precipitation estimates,
and, therefore, the study about the conversion of radar echo to
precipitation intensity. According to the experience collected during
the project, as summarized in Gjersten et al. (2003), no generally-
accepted methodology for radar-derived precipitation estimates is
available and many different approaches for improving the
quantitative precipitation estimates of weather radar are in use or
presently tested in different EU-countries.
The current and most applied methods of radar (R) adjustment
usually rely on rain gauge data (G), although different hypothesis
and aggregation procedures have been tested within this general
framework, as illustrated in Krajewski (1987), Borga et al. (2002),
Todini (2001) Seo (1998), and several others. Based on the
definition of Michelson (2003), we concentrated on the gauge-radar
adjustment problem, which involves any procedure whereby the
characteristics of radar data are modified such that they correspond
as much as possible with the measured quantity collected by rain
gauges. According to these recent studies, a comprehensive
statistical analysis and comparison of weather radar estimates and
rain gauges data were carried out.

5.1.2 Study area


The study area is situated upstream to the Lake Maggiore, on the
border between Switzerland and Italy. The area (6599 km2) is
mountainous, with elevation ranging from 200 m to about 3500 m
a.s.l., particularly prone to heavy rainfall and flood events.

5.2
Chapter 5 - Investigation of the radar rainfall estimates
and of the temperature interpolation method

5.1.3 Radar data


The study area is covered by the Mount Lema weather radar
operated by MeteoSwiss (see Figure 5.1). The C-band Doppler
weather radar was conceived to work continuously in an unattended
operational environment and to supply high quality data even in
presence of severe ground clutter. With a spatial resolution of the
volume element of 1 km3, 20 different elevations are scanned every
5 minutes and finally converted in 16 reflectivity levels. A best
estimate of the precipitation rate, called RAIN, containing the first
30 minutes precipitation intensity was used as data set for all the
analyses. Variability in time and space of the vertical profile of
reflectivity (VPR), issues related to beam geometry, ground clutter,
anomalous propagation, and visibility, stability of the electronic
calibration of the radar system are all already considered during the
processing chain up to the RAIN product. The values of reflectivity
are finally transformed by MeteoSwiss in rainfall intensity by using
the standard Z-R relationship:

Z = A · Rb [eq. 5.1]

where Z is the reflectivity [mm6/m3], R is the corresponding rainfall


intensity [mm/h]; A=316 and b= 1.5 are the parameters estimated
from Joss et al. (1997) for whole Switzerland.

5.1.4 Rainfall Data


Hourly rainfall measurements from 35 stations and 3 different
meteorological networks (see figure 5.1) are collected in real-time:
the network of the Institute of Earth Sciences (IST), the network of
Piemonte Region (RP) in Italy, and the network of MeteoSwiss (MS).
A data quality control is directly performed by MeteoSwiss and a
plausibility check is carried out, before the data are transferred into
the database. Rainfall data and radar estimates were available for
the whole Julian year 2002.

5.3
Chapter 5 - Investigation of the radar rainfall estimates
and of the temperature interpolation method

22

F
$
Legend 27
513 36 "
Rain Gauges
$ "39 $ 13 43
"21
network 18 $ $
" IST ! $
19
$ MS 53 2 "
! RP $ !
k
j !52 !14 11
7 34$ 29
Radar
24
DEM 37! 17 " $ $
16
Value
! ! ! !15
54 !1 50 Monte Lema
High : 4572.000000 28 ! k
j 25
$ 10 $
! ! 35
42
Low : 193.000000 ! 33 !
0 5 10 20 Kilometers ! !47 46
$

Figure 5.1 – Radar and rain gauge networks

In table 5.1 some basic statistics of the two data sets are illustrated.

Radar Rain Gauges


Observations No. 251068 251068
Mean 0.1 0.26
Variance 0.33 1.4
Min 0 0
Max 50 58.2
Table 5.1 – Basic statistics of the data sets in mm/h.

5.1.5 Preliminary analysis


A preliminary analysis of the data samples has been carried out, in
order to (i) verify absence of anomalies, (ii) detect presence of
outliers in the radar images, and (iii) homogenize the different
rainfall data sets.
Persistent anomalies have been excluded by verifying that none of
the rain gauges falls into a cell where the reflectivity in the RAIN
product exceeds 0.16 mm/h in at least 20% of the time in clear sky
conditions. Gross errors are identifiable by considering a 3x3 pixel
grid centered on the i-th pluviometer and by calculating, for each
time step, the mean value µ and the median m of the 9 observations.
By comparing the two values, the difference d=µ-m confirms the

5.4
Chapter 5 - Investigation of the radar rainfall estimates
and of the temperature interpolation method

absence of many outliers in the radar estimates corresponding to


the rain gauge positions (average=0.00, median 0.01, min.=-7.67,
max.=9.00 in mm/h).
The homogenization of the rainfall data were performed throughout
(i) a temporal synchronization of the rainfall data collected from
different networks, and (ii) a reclassification procedure. In order to
compare radar estimates with rain-gauge measurements, the
rainfall data were classified according to the boundaries of the
radar classes (3dBZ amplitude), as provided by MeteoSwiss.
To each class, for both radar and rain-gauge, the central value of
the class was finally associated. Any further analysis will refers to
this data set, whereas error sources not considered in RAIN were at
the moment neglected.
Finally, I assumed that gauge measurements are appropriate for
each gauge location and they are therefore referred to as the
“ground truth”. This is only partially true, since precipitation
measurements are affected by different errors.
Nevertheless Lombardo and Staggi (1998) showed that the
systematic error is finally a bias, which could be corrected using a
simple calibration curve. Additionally to the gauge measurement
errors, the question of using point measurements in representing
areal rainfall of a radar pixel has been emphasized by many authors
(see, e.g., Chumchean et al., 2003).

5.1.6 Rainfall detection analysis


As first test of the radar reliability, the simple detection of a rainfall
event, without any consideration of its magnitude was carried out,
by considering the binomial process rain no-rain, both for gauges
and radar. The relative concordance and discordance frequency of
the four possible cases are summarized in table 5.2.
A high agreement level (over 90%) is achieved in the case of both
rain and no-rain cases. Furthermore, referring the gauges as
ground-truth, the detection of a no-rain event by the radar is very
reliable, not only in the frequency (97%), but also in terms of

5.5
Chapter 5 - Investigation of the radar rainfall estimates
and of the temperature interpolation method

intensity, since in the empirical cumulative distribution function of


no-concordance cases only 0.66% correspond to error higher than 2
mm/h. The detection of a rainfall event is on the other side more
uncertain, with an error by the radar in 47% of the cases.

RAIN GAUGE
rain no-rain tot
rain 16218 6482 22700
RADAR
no-rain 18050 210318 228368
tot 34268 216800 251068
Table 5.2 – Four cell contingency table, gauges and radar

5.1.7 Rainfall intensity analysis


By visually comparing rain-gauge measurements (named RG) and
radar estimates (named RR), a tendency of the radar to
underestimate was evident. A seasonal analysis on the conditional
sample with RR>0 and RG>0 (herein referred to as “active-set”),
aiming at distinguish different characteristics of the rainfall events
(advective events in the cold season, convective events in the hot
season) confirms this tendency. The standard deviation of RR is
smaller than the one observed by RG and seasonally variable.
Moreover the values of D= RG - RR on the “active-set” are bias
affected since their mean and median are strongly different
(average=1.69, median=0.96, min.=-48.00, max.=56.00 in mm/h).

5.1.8 Spatial correlation analysis


A spatial correlation analysis of rainfall measurements was
undertaken, aiming at identifying spatial properties of the rainfall
field. This would allow radar adjustment with gauge data, e.g. by
Kriging or Cokriging procedures as in Krajewski (1987) or Todini
(2001).
The empirical semivariogram on the“active-set” was calculated as
follow:

5.6
Chapter 5 - Investigation of the radar rainfall estimates
and of the temperature interpolation method

{ yt ( si ) − yt ( s j )} with: 5k ≤ si − s j ≤ 5( k + 1)
35
1
γ ( hk ) = ∑ ∑
2

2N ( hk ) t si , s j =1 [eq. 5.2]

where hk=[5k, 5(k+1)] km, γ(hk) is the mean semivariance in the


range hk, N(hk) is the corresponding number of observed pairs in
the range hk, yt(si) is the rainfall observation at gauge si at time t.
Both radar and gauges display a phenomenon with similar
correlation length (sill), but their γ(hk) values, as illustrated in figure
5.2, present a non constant ratio among the empirical variograms
calculated at different months.
The different spatial correlation of the rainfall field, as observed by
gauges and radar, prevent us from directly and trivially integrating
the two datasets into a unique geostatistical model.

RG
R R
RR
RR
RG
R RR
RR
G G

Figure 5.2 – Monthly semivariograms for RG and RR


(June on the left and December on the right).

5.1.9 Linear and functional correlation analysis


By using linear correlation analysis a functional relationship
between RR and RG was investigated. The global correlation index
is extremely low (ρ =0.324), and the distribution of the single ρ
values at each gauge does not reveal any spatial pattern, therefore
indicating that a linear radar vs. rain gauges adjustment procedure
can not be implemented. A more general functional dependence was
thus investigated.
A general dependence of two variables Y and X can be estimated by
means of a functional index η2Y, defined according to Sansò (1996):

5.7
Chapter 5 - Investigation of the radar rainfall estimates
and of the temperature interpolation method

σ Sy2
η ( x, y ) = 2
2
Y [eq. 5.3]
σY

where: σ2SY = E{(E{Y|X=x}-μY)2}, σ2Y = E{(Y-μY)2}.


The index ranges from 0 to 1; η2Y = 1 if and only if Y is a function of
X, and η2Y = 0 if X and Y are stochastically independent. By
comparing this value with ρ a certain, but not satisfactory,
functional dependency can be identified. Taking into account the
characteristics of the data, particularly: the radar data availability
(rainfall estimates of the first 30 minutes of each hour), and the
physical behaviour of the rainfall phenomenon in narrow and deep
alpine valleys, a 3 hours moving average operator was applied to
RR.
The correlation between radar and rain gauges significantly
increased suggesting to look for a new parameters estimate of the
Z-R relation using ZRm (3 hours moving average reflectivity) and RG.
The results are summarized in figure 5.3.

1 η(ZR,RG)
η(ZRm,RG)
0.9
functional correlation index square root

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0
1 2 3 7 10 11 13 14 15 16 17 18 19 21 22 24 25 27 28 29 33 34 35 36 37 39 42 43 46 47 50 51 52 53 54
Rain Gauges

Figure 5.3 – Square root of the functional correlation index for different
gauges and radar.

5.1.10 Z-R relationship adjustment


According to the previous analysis, α and β of the inverse of the
radar equation

5.8
Chapter 5 - Investigation of the radar rainfall estimates
and of the temperature interpolation method

RG = α ⋅ ΖRmβ [eq. 5.4]

were fitted for each station, by non-linear least-square estimation


and further converted into the commonly used A and b (see figure
5.4 and equation 5.1). The result of the fitting shows that A values
mostly range from 0 to 50 and b values range from 2 to 4 and
therefore they greatly differ from those operationally used by
MeteoSwiss. A high spatial variability can be also observed.
By using an outliers identification procedure (Grubbs’ Test) 3
stations (No. 18, 36, 54) were excluded from further analysis. The
result of the Grubbs’ Test fairly agrees with consideration related to
the local meteorological and topographical conditions of the
identified critical sites. By omitting these stations, a general Z-R
relation for the whole data set was estimated; nevertheless this
approach is still unsatisfactory, since the residual variance curve is
higher than the one computed at each gauge (see as example
figures 5.5 and 5.6).

-200 -150 -100 -50 0 50 100 150 200


54
53
52
51
50
47
46
43
42
39
37
36
35
34
33
29
28
27
25
24
22
21
19
18
17
16
15
14
13
11
10
7
3
2
1

20 15 10 5 0 -5 -10 -15 -20


b_Hot b_Cold b A_Hot A_Cold A

Figure 5.4 – Estimated Z-R parameters in different seasons (Hot and


Cold) and all over 2002.

5.9
Chapter 5 - Investigation of the radar rainfall estimates
and of the temperature interpolation method

Figure 5.5 – Regression curve (reg), residual variance curve (var), and
fitted model (fit) on gauge No. 1 data

Figure 5.6 – Regression curve (reg), residual variance curve (var), and
fitted model (fit) on the whole data set (right).

5.10
Chapter 5 - Investigation of the radar rainfall estimates
and of the temperature interpolation method

5.1.11 Conclusions of the validation of radar rainfall estimates


It has been shown that, in the study area, the MeteoSwiss RAIN
product:
 can be used as reliable tool for no-rain detection;
 underestimates rain average and variance values in comparison
with gauge observations;
 shows different spatial correlations compared to those usually
observed from gauge data.
To achieve a reliable rainfall intensity map, a recalibration of radar
measures is needed.
The use of a moving average operator on radar reflectivity
observations generates a high non linear functional correlation
between the consequent radar data and the rain gauges
measurements. This functional dependence could be modelled by
means of locally recalibrated Z-R relations.
The strong space-wise and time-wise variability for the computed
parameters of the Z-R relation suggests that a local fit and a further
interpolation, instead of using constant A and b values for the whole
area, could be a methodology to achieve a reliable rainfall intensity
map. Anyway the studied product (RAIN) has been shown to be not
jet mature for quantitative precipitation estimations and therefore a
classical approach has been chosen.
One question still remains open: is it higher the radar error in
magnitude estimates or the interpolation error in spatialize point-
wise observation? No answer can be found at the moment because
of the lack of any reliable measures.
It could be argument for further research to compare hydrological
models simulations (estimated river flow discharges) obtained with
gauges inputs and with radar estimates inputs to the observed
discharges, in order to verify the reliability of radar rainfall
estimates in hydrological modelling.

5.11
Chapter 5 - Investigation of the radar rainfall estimates
and of the temperature interpolation method

5.2 Investigation of interpolation methods for


temperatures maps derivation
Surface air temperature is an important meteorological input in
most formulas for the evapotranspiration and snowmelt calculation.
Therefore, an accurate interpolation of temperature is necessary.
Due to the lack of numerous observations, the topography
dependence of this variable has been often taken into account in
order to improve the spazialization of the sampled points.
Different interpolation schemes were evaluated in literature using
the leave-one-out cross-validation method (see e.g. Collins and
Bolstad, 1996).
The interpolation schemes often applied are:
1. constant lapse rate;
2. inverse distance weighting (IDW) without consideration of
elevation effects;
3. ordinary Kriging without consideration of elevation effects;
4. hybrid scheme that combined a constant lapse rate with ordinary
Kriging;
5. elevationally detrended kriging approach, which is a combination
of regression against elevation and ordinary Kriging (Li et al.,
2003).

After an investigation of the different techniques and the influence


on the temperature of various parameters, the elevation (Z) was
found as the most significant one. As balance between accuracy and
simplicity three models was tested:

1. MODz – a simple linear relationship between the temperature (T)


and the height (Z):
T = aZ + b [eq. 5.5]

2. MODz2 – a 2nd order polynomial relationship between


temperature and height trying to take into account the

5.12
Chapter 5 - Investigation of the radar rainfall estimates
and of the temperature interpolation method

phenomenon of the inversion temperature profile:


T = aZ2 + bZ + c [eq. 5.6]

3. MODxyz – a 3rd order polynomial in X and Y with an additional


term in Z to consider temperature attenuation with elevation:
T = b + b1X + b2Y + b3X2 + b4Y2 + b5XY
+ b6X3 + b7Y3 + b8X2Y + + b9XY2 + b10Z [eq. 5.7]

This three methods were validated by using the leave-one-out cross-


validation procedure: one observation at time is leaved out from the
interpolation; the omitted observation is then used to compute the
residual between its value and the one generated, by means of
interpolation, without its contribution.
For each map (for each hour) it is possible to evaluate the MBE
(mean bias error) and the MAE (mean absolute error) following the
equation 5.8 and 5.9:

∑n Es i − Eoi
[eq. 5.8]
MBE = ∑ Esn − Eo
i =1
i i
MBE = i =1

n
n

∑n Es i − Eoi
MAE = ∑ Esn − Eo
i =1
i i [eq. 5.9]
MAE = i =1

where:
n = number of observation, Es = estimated value, Eo = omitted
observation, i = observation index.
Applying this procedure for multiple time series it is possible to
evaluate a series of these indexes that indicates the goodness of the
interpolations.
The data set for the cross-validation procedure was composed by a
total of 251'959 observations coming from 32 stations (see table

5.13
Chapter 5 - Investigation of the radar rainfall estimates
and of the temperature interpolation method

5.3) with hourly sampling lap and referring to the whole year 2002.
This approach has been used to evaluate the performances of the
MODz, MODz2 and MODxyz model for temperature interpolation.

Nid Station Name Descr Est Nord Quota


1 Alpe di Cheggio ant 652171 104165 1510
2 Alpe Devero bac 663117 129561 1634
10 Candoglia can 676102 91818 201
11 Cimetta cim 704370 117515 1672
13 Comprovasco com 714998 146440 575
14 Crodo cro 668170 119610 560
15 Cursolo Orasso cur 687181 105678 940
16 Domodossola dom 666627 105974 252
17 Druogno dru 676289 109454 831
18 Piano dei Camosci for 670747 142739 2450
22 Guetsch gue 690140 167590 2287
24 Lago Paione Sup. lpa 657981 114162 2269
25 Lugano lug 717880 95870 273
28 Passo del Moro mac 641739 93894 2820
29 Locarno-Magadino mag 711160 113540 197
30 Monte Matro mat 714260 140940 2171
33 Mottarone mot 678326 81222 1302
34 Locarno-Monti olm 704160 114350 366
35 Pallanza pal 685952 86297 211
36 Piotta pio 694930 152500 1007
37 Pizzanico piz 657116 108101 1142
39 Robiei roe 682600 144075 1898
42 Sambughetto smb 667557 84089 800
43 S. Bernardino sbe 734120 147270 1639
46 Stabio sta 716040 77970 353
47 Someraro str 683252 82060 430
50 Alpe Mottac tro 674650 101342 1690
51 Ulrichen ulr 666740 150760 1345
52 S. Domenico var 658120 122068 1300
53 Visp vis 631150 128020 640
54 Zermatt zer 624350 97550 1638
55 Lodrino ldr 719650 128500 261
56 Generoso gen 722250 128500 1608
57 Cevio cev 689700 130550 416

Table 5.3 – stations used in temperature interpolation methods analysis

Based on this data set the MBA and MBE where calculated at each
hour: in table 5.4 a statistic summary of these indexes is presented.
The model MODxyz results in being too inaccurate with a mean

5.14
Chapter 5 - Investigation of the radar rainfall estimates
and of the temperature interpolation method

MAE of 4.3°C while MODz and MODz2 results to be fairly good with
a mean MEA values of 0.42°C.

MODxyz MODz MODz2


MAE max 195.589 5.8153 5.8321
min 0.418 0.4275 0.4405
mean 4.368 1.5298 1.5219
median 1.778 1.3532 1.3451
MBE max 56.09081 0.243848 0.70255
min -56.38430 -0.284020 -0.18990
mean -0.09006 0.001829 0.01669
median 0.09309 0.001834 0.01203

Table 5.4 – MAE and MBE statistics for the 2002 temperature (°C) time
series

Figure 5.7 shows the MAE behaviour along the whole year 2002
with the three evaluated methods.

5.15
Chapter 5 - Investigation of the radar rainfall estimates
and of the temperature interpolation method

Figure 5.7 – temporal distribution of MAE with three interpolation


methods; the horizontal lines represent the mean value.

Also figure 5.8 shows at a given time (2002-01-14 00:00) that MODz
has better fitting than MODxyz.

5.16
Chapter 5 - Investigation of the radar rainfall estimates
and of the temperature interpolation method

Figure 5.8 – observations vs. estimations: with the “o” symbol the
MODxyz prediction and with the “+” symbol the MODz prediction.

A further confirm of that is given by the visualization of these maps,


as shown in figure 5.9.

5.17
Chapter 5 - Investigation of the radar rainfall estimates
and of the temperature interpolation method

Figure 5.9 – maps generated with different interpolation methods. In


clock wise order from the top left map: weighed inverse distance,
MODxyz, MODz and MODz2.

The cross-validation shows how the two interpolation methods


MODz2 and MODz are very close in generating interpolation errors;
therefore the simplest one has been selected.
The implemented procedure is composed of three steps:
 querying of the MNSD database and extraction of the
temperatures and relative elevation of observation;
 regression analysis and calculation of the regression parameters;
 application of the estimated relationship to the elevation map by
means of map algebra;
As shown in frame 5.1 a pearl script executes the query (based on a
system variable defining the correct timestamp) and the regression
analysis returning the parameters A and B used for mapalgebra by
the r.mapcalc GRASS command.

5.18
Chapter 5 - Investigation of the radar rainfall estimates
and of the temperature interpolation method

L1> # query and calculation of reg. parameters


L2> COEFF=`perl temp_regression.pl`
L3> A=`echo $COEFF | awk -F" " '{print($1);}'`
L4> B=`echo $COEFF | awk -F" " '{print($2);}'`
L5> r.mapcalc temp.200503151300 =$A+$B*dem
Frame 5.1 – temperature map derivation procedure.

5.19
Chapter 6 – Case study: the Verzasca basin

6. Case study: the Verzasca basin


In order to test the developed procedures and modules, they have
been applied to the Verzasca basin as a case study.
The analysed period is the August 2004, where same short showers
and one heavy rainfall event occurs.
After a general introduction to to the geographic region of the
Verzasca basin, the most significant produced maps have been
analysed and validated by means of: verified realible spatial
representation, and comparisons with other available data and “in
situ” observations.

6.1 General characteristics


The Verzasca valley, is located in southern Switzerland, Canton
Ticino (see figure 6.1). The Verzasca river flows from the Alps to the
Lake Maggiore and has a length of 34 Km with a mean discharges of
1 m3/s that, during floods events, can easily rise up to 250 m3/s. The
drained area has an extension of 218 Km 2 and registers a mean
annual precipitation of 1976 mm with a mean of 129 rainy days a
year. The basin shows the typical Alpine characteristics with its
high elevation ranges (from 193 m to 3596 m a.s.l.), narrow valley
(2 Km wide), and high mean slope (33.93°).

Figure 6.1 – Verzasca valley localization in Switzerland.

6.1
Chapter 6 – Case study: the Verzasca basin

Figure 6.2 and 6.3 show a picture and a satellite image of the
valley.

Figure 6.2 – picture of the Verzasca valley at Ponte Brolla

Figure 6.3 - Satellite image of the Verzasca valley


(source http:\\maps.google.com)

6.2
Chapter 6 – Case study: the Verzasca basin

6.2 Projection and region properties


The projection used in the case study, following the proj
structure(www.remotesensing.org/proj/gen_parms.html, April 2006)
is:
 name : Swiss. Obl. Mercator
 datum : ch1903
 towgs84 : 674.253,15.053,405.324
 proj : somerc
 ellps : bessel
 a : 6377397.1550000003
 es : 0.0066743722
 f : 299.1528128000
 lat_0 : 46.9524055560
 lon_0 : 7.4395833333
 k_0 : 1.0000000000
 x_0 : 600000
 y_0 : 200000
 unit : meters
 units : meters
 meters :1

The region of application, that will be maintained during all the


processing phases and saved with the name of Verzasca_reg
(g.region save=Verzasca_reg;) is identified by the following
coordinate of the limits:
north=140636.61458333; south=115129.97395833;
west=695548.32142829; east=715255.35714256.
The resolution in north-south and east-west has been fixed at 100 m
and the rows number and columns number are respectively 255 and
197, for a total of 50235 cells.

6.3
Chapter 6 – Case study: the Verzasca basin

6.3 Topography
The topography is described by the Swiss elevation model DHM25
at 25 meter resolution (derived from topographic maps at scale
1:25'000), with a vertical accuracy of 2 m for the Ticino area. A 3D
view of the basin, generated with the GRASS nviz command, is
shown in figure 6.4.
The DTM over the basin area has minimum value of 193 m,
maximum value of 3596 m, and mean value of 1608 m a.s.l.

NORTH

Figure 6.4 – 3D view of the Verzasca basin.

6.4
Chapter 6 – Case study: the Verzasca basin

6.4 DTM derived maps


As discussed in paragraph 4.6, in order to calculate the runoff
discharges some maps has to be derived from the elevation model.
The DTM has been resampled at 100 m. cell size by mean of the
r.resample GRASS command, this because the other informations
(e.g. Landuse and soil map) are available with this resolution. In this
section the application of the procedures described in chapter 4 and
the generated map are shown and discussed.

6.4.1 Filled DEM


The digital elevation model has been modified in order to remove
sinks, that generally would not permit the continuous flow over the
basin. This map can be either generated with the r.fill.dir or the
r.terraflow commands (see figure 6.5). The second one has a more
sophisticate algorithm that better solves problems of flow in lowland
areas and thus it has been preferred (r.terraflow elev=dem
filled=dem_fill dir=direction swatershed=watershed
accumulation=accu tci=tci;).

Figure 6.5 - Filled DEM [m] generated with the r.terraflow algorithm.

6.5
Chapter 6 – Case study: the Verzasca basin

Unfortunately the maps of flow direction and flow accumulation that


can be generated by r.terraflow cannot be used by the hydroFOSS
model.
In fact the r.terraflow command solve the flow directions with a so
called “D8-MFD” algorithm (Tarboton, 1997) that spread the flow
direction among all the neighbour cells with lower elevation.
While, as discussed in chapter 4.6, for sake of simplicity the
h.hydrofoss.runoff model solves the runoff only in one of the four
cardinal direction, and therefore the flow direction map has to be
extracted with the so called “D4-SFD” algorithm.
For this reason the r.watershed command has been chosen (as
described in the following section) and a filled DEM is no more
necessary because the used algorithm directly override sink
problems.
In the next pages the command applied for this purpose has been
described.

6.4.2 Stream, basin, flow accumulation, and flow direction maps


Following the procedure reported in chapter 4.6 all these maps
have been created with the r.watershed command:
r.watershed -4 elevation=dem_fill threshold=100
accumulation=ACCUMUL drainage=DIRECTION basin=BASIN
stream=STREAM;.
The threshold value of 100 has been chosen among some test values
because it is the one that, in the Verzasca basin case, permits to
extract a stream map in good agreement with the “observed” river
layer from the Swiss “vector25” maps of the Canton Ticino: the
fitness of the two maps is represented in figure 6.6 and 6.7.
In figure 6.8, 6.9, and 6.10 the basin, the flow accumulation, and
the flow direction maps are shown.
Note that the basins that are not completely defined within the
region limits cannot be solved in the runoff process, because the
incoming flow could be unknown.

6.6
Chapter 6 – Case study: the Verzasca basin

Figure 6.6 – extracted river (white) and “observed” river (blue)


(threshold=100).

Figure 6.7 – zoom of extracted river (white) and “observed”


river (blue) (threshold=100).

6.7
Chapter 6 – Case study: the Verzasca basin

Figure 6.8 – Basins map calculated with the r.watershed


command (threshold=100).

The flow accumulation map defines the number of cells that drain
through each cell.
The r.watershed command extracts a map whose absolute cell
values represent the amount of upland cells plus one. Negative
numbers indicate that those cells possibly have surface runoff from
outside of the current geographic region.
Thus the not completely defined basins will have in the flow
accumulation map (figure 6.9) some cells with negative values and
will not be solved in the h.hydroFOSS.runoff module.
It is therefore really important to define the correct region limits
containing all the basins that have to be processed.

6.8
Chapter 6 – Case study: the Verzasca basin

Figure 6.9 – Flow accumulation map of the Verzasca region.

The flow direction map represents the drainage direction.


The r.watershed command extracts a map whose absolute cell
values, multiplied by 45 will give the direction in degrees that the
surface runoff will travel from that cell. Negative values indicate
that surface runoff is leaving the boundaries of the current
geographic region.
Because the r.watershed has been run with the -4 flags, that
indicate the usage of the “D4” algorithm, only the absolute values 2
(90°, North), 4 (180°, West), 6 (270°, South), and 8 (360°, East) are
present (see figure 6.10) while some negative values can be noted
in correspondence of the region borders.

6.9
Chapter 6 – Case study: the Verzasca basin

Figure 6.10 – Flow direction map for the Verzasca region.

As the basin is not too large and our computer power is elevated
(one cell requires 132 byte of memory) we decided to aggregate the
basins; this operation can be done with two commands:
 r.statistics in order to extract a map with the minimum
accumulation value for each basin: “r.statistics base=BASIN
cover=ACCUMUL method=min output=min_accumul;”
 r.mapcalc in order to generate two main basins, one completely
defined in the region and one with all the other basins: “
r.mapcalc 'VerzBasin=if(@min_accumul>0,1,2)'”

The new generated basins map (see figure 6.11) will be used in the
h.hydroFOSS.runoff command.
For the runoff processing it will require 2.75 MB (21'837 cells * 132
byte = 2'882'484 byte).

6.10
Chapter 6 – Case study: the Verzasca basin

Figure 6.11 – Accumulated basins for runoff calculation.

6.11
Chapter 6 – Case study: the Verzasca basin

6.4.3 Slope map


The slope map can be derived by running the r.slope.aspect
command with the filled DEM as input map:
“r.slope.aspect elevation=dem_fill slope=slope;”
Figure 6.12 shows the generated map (masked with the VerzBasin
map), while the table 6.1 shows same basic statistics.

Figure 6.12 – Slope map [deg] generated with r.slope.aspect algorithm.

Average 36.72
Stdv 10.44
Min 0.1
Max 66.34
Table 6.1 – statistical values for slope map in degrees;

6.12
Chapter 6 – Case study: the Verzasca basin

6.5 Landuse
The used landcover map is the “Arealstatistik 1972“, derived from
topographic maps. It was elaborated from the Swiss Federal Office
of Statistic which classified the landuse in 12 categories over a grid
with resolution of 100 m according to the dominance principle that
assigns to the cell the most extended category.
Ten of these categories are present in the Verzasca basin, while one
additional class (river) has been added, for hydrological modelling
porpoises, superimposing the extracted stream map.
In fact, due to the dominance principle, the water class do not
represent all the cell where the river is present while the
characterization of the river properties is required from the
r.hydrofoss.init module (see section 4.6.1).
Figure 6.13 shows the modified “Arealstatistik 1972“ map for the
test case area, while table 6.2 shows class statistics and figure 6.14
shows classes distribution over the entire basin where class with
small extension are not represented.

Figure 6.13 - modified “Arealstatistik 1972“ map over the Verzasca


basin.

6.13
Chapter 6 – Case study: the Verzasca basin

class classification area (Km2) cells Area %


1 water 2.93 293 1.34%
2 settlements 0.64 64 0.29%
3 pine forest 53.43 5340 24.47%
4 decidous tree 25.59 2557 11.72%
5 mixed forest 11.47 1146 5.25%
6 agriculture 0.11 11 0.05%
7 grass 56.08 2001 9.17%
8 bushes 20.02 5605 25.68%
9 rock 37.40 3738 17.13%
10 horticulture 0.01 1 0.01%
11 river 10.83 1082 4.95%

Tbale 6.2 – landuse classes statistics for the Verzasca basin.

Landuse classes distribution

river water settlements


1.34%
horticulture 4.95% 0.29%
0.01%

rock pine forest water


17.13% 24.47% settlements
pine forest
decidous tree
mixed forest
agriculture
grass
bushes
rock
decidous tree horticulture
11.72%
river
bushes
25.68% mixed forest
5.25%
grass agriculture
9.17% 0.05%

Figure 6.14 – landuse classes distribution in the Verzasca basin.

It is clear how the major landuse classes for the Verzasca basin are
the pine forests (class 3), the bushes (class 8), and the rocks (class
9).

6.14
Chapter 6 – Case study: the Verzasca basin

6.6 Soiltype
The soil map of the basin has been based on the Swiss soil data set
(known as “Bodeneignungskarte”), which includes information
about the suitability of soils for agricultural use at a resolution of
100 m. General information about soil dependent parameters was
retrieved from the so-called “Bodeneignungskarte” of Switzerland,
which describes soil attributes according to a proprietary
classification that aims at evaluating the agricultural attitude of
soils. The depths are therefore referred only to those soil layers that
are influenced by biological activities.
In figure 6.15 the soiltype maps for the Verzasca basin is shown,
while in table 6.3 and figure 6.16 information on the soiltype classes
distribution are represented.

Figure 6.15 – map of soiltypes found in the Verzasca basin


(derived from the “Bodeneignungskarte”).

6.15
Chapter 6 – Case study: the Verzasca basin

class classification area (m2) cells Area %


2 Lithosol predominant 130470536 13039 59.71%
3 Regosol eutric predominant 12887955 1288 5.90%
4 Ranker predominant 33080420 3306 15.14%
7 Debris and block areas 1030636 103 0.47%
8 Bare rock predominant 35762075 3574 16.37%
9 Sedimentary rock predominant 5273255 527 2.41%

Table 6.3 – soiltype classes statistics for the Verzasca basin.

Soiltype classes distribution


Sedimentary rock
Bare rock
predominant
predominant
2.41%
16.37%

Debris and block


areas
0.47%
Lithosol predominant
Regosol eutric predominant
Ranker predominant
Debris and block areas
Bare rock predominant
Sedimentary rock predominant
Ranker predominant
15.14%

Lithosol predominant
Regosol eutric 59.71%
predominant
5.90%

Figure 6. 16 – soiltype classes distribution in the Verzasca basin.

Lithosol predominant (class 2) is largely the most diffuse soil type in


the Verzasca area, and therefore it is expected that theirs derived
parameters will greatly influence the model simulations.

6.16
Chapter 6 – Case study: the Verzasca basin

6.7 Riverwidth
Not commonly adopted methodology to estimates the riverwidth
(rw) can be found in literature, thus a simple but logic procedure
has been used. With the hypothesis that the river width grows
following the drained area dimension, a linear factor growing with
the flow accumulation values has been estimated by means of:

Frw = [max(rw)-min(rw)]/[max(acc)-min(acc)] [eq. 6.1]

where:
max(rw) is the maximum riverwidth [m],
min(rw) is the minimum riverwidth [m],
max(acc) is the maximum flow accumulation at max(rw),
min(acc) is the minimum flow accumulation at min(rw),
Frw is the riwerwidth growing factor [m/accumulation];

After some “in situ” observations the river has been detected to
reach a maximum width of 10 m.
Observing that the maximum flow accumulation of the river's cell is
28500 and the minimum is 100 (this is the threshold applied to
derive the stream map) for the case study the used values are:
max(rw)=10, min(rw)=0, max(acc)=28500, and min(acc)=100
The resulting linear factor is therefore: Frw = 0.00035.
The Frw was then applied by means of map algebra to the flow
accumulation map and the riverwidth map was extracted (figure
6.17) by means of the following command:
r.mapcalc 'riverwidth=if(accumlation>10,accumulation*
0.00035,0)';

Furthermore the hydrological type map (figure 6.18) can be easily


extracted from the riverwidth map by means of map algebra with
the command:
r.mapcalc 'hydrotype=if(riwerwidth=0, 0, 1)';

6.17
Chapter 6 – Case study: the Verzasca basin

Figure 6.17 – riverwidth map for the Verzasca basin.

Figure 6.18 – hydrotype for the verzasca basin: 0=terrain, 1=river

The substantial correctness of the derived maps has been verified


by means of visual comparing with aerial photos of the Verzasca
basin.

6.18
Chapter 6 – Case study: the Verzasca basin

6.8 Model verification and running strategy


Known the basic maps (filled DTM, slope, soiltype, landuse,
riverwidth, hydrotype, flow accumulation, flow direction, and basin)
and defined the parameters tables (landuse_cat and soiltype_cat
tables, as defined in section 4.6.1 and shown in tables 6.4 and 6.5) it
is possible to run the simulation of the rainfall-runoff processes
during a given time interval.
The starting values has been selected as follow:
 Manning (surface roughness coefficient), estimated from the
literature (Chow, 1973) for each land cover class, and
subsequently adjusted to match the timing concentration towards
the drainage network;
 kappa_s (horizontal permeability), estimated for each soil class by
using values from the US Geological Survey (USGS) “South
Florida Information Access” (SOFIA) database
(http://sofia.usgs.gov/publications/wri/02-4050/hydcondsa.html,
2006);
 theta_s and theta_r (saturated and residual soil water content),
estimated by assigning for each soil type values taken from the
digital thematic soil map of Switzerland (Bodeneignungskarte,
see section 6.6);
 alpha (transmissivity law exponent), was fixed at 2.5, being the
value suitable for most of the soil as suggested in literature on
the basis of the application of the Topkapi model to other basins;
a lower value would imply a decrease in soil water drainage,
therefore an increase in the surface runoff;
 thik (soil thikness), derived from the “Bodeneignungskarte” as
an average value of the corresponding soil class in the soil map of
switzerland adjusted within 30 cm.;

In order to verify and calibrate the model, the full script was divided
in two parts:
 the first (h.hydroFOSS.pre) generates all the maps requested by
the flows computation (climatic maps, solar radiation maps, clear

6.19
Chapter 6 – Case study: the Verzasca basin

sky index map, global radiation maps, snow accumulation & melt
maps, evapotraspiration maps, canopy interception maps, and
available water for flows map);
 the second (h.hydroFOSS.cal) run and calibrates the runoff
model.

lu_id cat_name manning


1 lakes 0.8
2 settlements 0.01
3 pine forest 0.2
4 decidous tree 0.1
5 mixed_forest 0.15
6 agriculture 0.15
7 grass 0.18
8 bushes 0.09
9 rock 0.1
10 horticulture 0.15
11 rivers 0.04
Table 6.4 – landuse_cat table.

cat smb descr kappa_s teta_s teta_r alfa thik


2 l Lithosol 5.56E-005 0.524 0.141 2.5 1.1
3 re Regosol 1.10E-004 0.582 0.123 2.5 1.05
4 ra Ranker 1.78E-005 0.533 0.181 2.5 1.3
7 de Debris 5.56E-005 0.524 0.141 2.5 1.1
8 rk Bare rock 1.66E-004 0.506 0.123 2.5 1.1
9 se Sediment 1.31E-004 0.451 0.123 2.5 0.9
Table 6.5 – soiltype_cat table of the categories
found in the Verzasca Valley.

The division of the model in two different scripts, allows the user to
compute the raster series not affected by calibration only once and
to use these maps, already extracted, as inputs for the further
runoff running required by the calibration process. This solution

6.20
Chapter 6 – Case study: the Verzasca basin

allows to save time and computational power.


The raster series was hourly calculated all along the 2004 with
h.hydroFOSS.pre and verified:
 by displaying raster map series and visually verify the
correctness spatial component;
 by plotting sampled series at samples points and visually verify
the correct behaviour of the variables;
 by quantitatively compare annual estimated values with annual
observed values.

The calibration was run for the whole August 2004: the used
methods, described in chapter 4.8, was applied by comparing the
simulations with the observations at two river gauges stations
(Pincascia and Lavertezzo) that are shown in figure 6.19.
The parameters to be calibrated are those represented in the tables
landuse_cat and soiltype_cat, and their values can change during
the calibration process.

PINCASCIA

LAVERTEZZO

Figure 6.19 – Raiver gauges used in model calibration.

6.21
Chapter 6 – Case study: the Verzasca basin

6.9 Climatological maps


The processes described in section 4.1.2 was successfully applied,
and the hourly climatic raster series was generated.
As already discussed, the application of Thiessen tassellation do not
alters the observed values but just identify for each measure an
area of influence where the observation is uniformly applied.
This kind of interpolation, used for the rainfall, wind, and humidity
maps generation, produces maps with constant values that change
with abrupt artificial break lines. This representation unlikely
reproduce the reality and therefore a validation with other
information would not be meaningful.
For this reason a full validation of these maps with other data type
(e.g. remote sensing) cannot be performed. A simple check, that
verified the absence of null values and missing maps, validated the
generated series.
Furthermore, considering the general small number of climatic
measurements points (in this case study a maximum of 5 station
influencing the basin can be counted), a leave-one-out cross-
validation procedure will result in being inadequate due to the great
influence of every observation in the map generation: the omission
of one of the already limited interpolation points would greatly
affect the result.
From this point of view, the remotely sensed data, once
demonstrated their correctness in quantities estimation, could
greatly improve the representation of these fields.
Other improvements in the generation of these maps could be
achieved by taking advantage of dependencies from other variables
(like it has been done for the temperature maps interpolation in
relation with the quota, see chapter 5.2).
Further investigation on each of these climatic variable (wind
velocity, humidity, and rainfall) should be carried on in future
studies.
Validation and investigation on temperature map series has been
discussed in chapter 5.2.

6.22
Chapter 6 – Case study: the Verzasca basin

In figures 6.20-27 some examples of rainfall maps (figure 6.20-21),


wind velocity maps (figure 6.22-23), temperature maps (figure 6.24-
25), and humidity maps (figure 6.26-27) calculated at different
instants are shown.
With exception for the temperature maps, where a topographic
variable is considered, these figures reveal the already mentioned
presence of artificial break lines. This behaviour is going to
influence all the maps derived by the usage of these climatological
maps, as can be easily noted looking at the map derived with the
snow module (Figures 6.34, 6.35, 6.36), the interception module
(Figures 6.40, 6.41, 6.42) and the runoff module (Figures 6.53, 6.54,
6.55).
This kind of approach (Thiessen tasselation) for climatological maps
extraction has been widely used mainly for two reason:
1. hydrologists are mostly interested in computing variables (mainly
discharges) for an outlet, and therefore, in this way, they
consider a sort of cumulative values (in space, zime and
processes) that hardly suffer from those artificial boundary;
2. proven methodology for distributed observations (quantitatively
valuables) or more dense samples are missing, therefore no
sensible advantage can be achieved without exhaustive
researches.
Anyway, still remaining known errors in the spatial description of
same veriables, as shown in the further sections, the general
behaviour of the basin can still be simulated.

6.23
Chapter 6 – Case study: the Verzasca basin

2.3 0.3

0.1

0.0

Figure 6.20 – Rainfall map [mm/h] for the 19/08/2004 at 18:00

0.3

0.0

0.8

Figure 6.21 – Rainfall map [mm/h] for the 20/08/2004 at 18:00

6.24
Chapter 6 – Case study: the Verzasca basin

3.8

0.2

0.8

2.1

1.4

Figure 6.22 – Wind velocity [m/s] map for the 19/08/2004 at 18:00

11.1

1.1

1.6

5.3

0.4

Figure 6.23 – Wind velocity [m/s] map for the 20/08/2004 at 18:00

6.25
Chapter 6 – Case study: the Verzasca basin

Figure 6.24 – Temperature [°C] map for the 20/08/2004 at 18:00 (day)

Figure 6.25 – Temperature [°C] map for the 21/08/2004 at 01:00 (nigth)

6.26
Chapter 6 – Case study: the Verzasca basin

86
100

97

86
84
Figure 6.26 – Humidity [%] map for the 19/08/2004 at 18:00

85
98

81

83

73

Figure 6.27 – Humidity [%] map for the 20/08/2004 at 18:00

6.27
Chapter 6 – Case study: the Verzasca basin

6.10 Watershed characterization maps


These maps has been extracted from literature values with the
commands h.seasonal.par and h.monthly.par described in section
4.1.3. The used monthly parameters table (table 6.6) and seasonal
parameters table (table 6.7) were derived from a WaSiM model
application in the Alpine region (Schulla and Jasper, 1999).
Since the albedo is time independent, in table 6.6, the same values
are repeated through columns.

PAR RASTER CAT JEN FEB .. .. .. .. .. .. SET OCT NOV DEC


alb landuse_cat 1 0.05 0.05 .. .. .. .. .. .. 0.05 0.05 0.05 0.05
alb landuse_cat 2 0.1 0.1 .. .. .. .. .. .. 0.1 0.1 0.1 0.1
alb landuse_cat 3 0.12 0.12 .. .. .. .. .. .. 0.12 0.12 0.12 0.12
alb landuse_cat 4 0.17 0.17 .. .. .. .. .. .. 0.17 0.17 0.17 0.17
alb landuse_cat 5 0.15 0.15 .. .. .. .. .. .. 0.15 0.15 0.15 0.15
alb landuse_cat 6 0.25 0.25 .. .. .. .. .. .. 0.25 0.25 0.25 0.25
alb landuse_cat 7 0.25 0.25 .. .. .. .. .. .. 0.25 0.25 0.25 0.25
alb landuse_cat 8 0.2 0.2 .. .. .. .. .. .. 0.2 0.2 0.2 0.2
alb landuse_cat 9 0.12 0.12 .. .. .. .. .. .. 0.12 0.12 0.12 0.12
alb landuse_cat 10 0.25 0.25 .. .. .. .. .. .. 0.25 0.25 0.25 0.25
alb landuse_cat 11 0.05 0.05 .. .. .. .. .. .. 0.05 0.05 0.05 0.05
rsc landuse_cat 1 20 20 .. .. .. .. .. .. 20 20 20 20
rsc landuse_cat 2 100 100 .. .. .. .. .. .. 100 100 100 100
rsc landuse_cat 3 80 80 .. .. .. .. .. .. 55 75 80 80
rsc landuse_cat 4 100 100 .. .. .. .. .. .. 65 85 100 100
rsc landuse_cat 5 90 90 .. .. .. .. .. .. 60 80 90 90
rsc landuse_cat 6 80 80 .. .. .. .. .. .. 65 75 90 90
rsc landuse_cat 7 90 90 .. .. .. .. .. .. 60 70 90 90
rsc landuse_cat 8 80 80 .. .. .. .. .. .. 55 70 80 80
rsc landuse_cat 9 250 250 .. .. .. .. .. .. 250 250 250 250
rsc landuse_cat 10 100 100 .. .. .. .. .. .. 60 80 100 100
rsc landuse_cat 11 20 20 .. .. .. .. .. .. 20 20 20 20
Table 6.6 – monthly parameters table.

The used seasonal parameters, of table 6.7, were referred to the


four reference days of the year: DAY1=110, DAY2=150, DAY3=250,
and DAY4=280. Also here some values are repeated trough
columns: they are clearly no time dependent.

6.28
Chapter 6 – Case study: the Verzasca basin

PAR RASTER CAT DAY1 DAY2 DAY3 DAY4


LAI landuse_cat 11 1 1 1
LAI landuse_cat 21 1 1 1
LAI landuse_cat 38 12 12 8
LAI landuse_cat 4 0.5 8 8 0.5
LAI landuse_cat 52 10 10 2
LAI landuse_cat 61 5 3 1
LAI landuse_cat 72 4 4 2
LAI landuse_cat 83 5 5 3
LAI landuse_cat 91 1 1 1
LAI landuse_cat 10 0.5 5 5 0.5
LAI landuse_cat 11 1 1 1 1
vegh landuse_cat 1 0.01 0.01 0.01 0.01
vegh landuse_cat 2 10 10 10 10
vegh landuse_cat 3 10 10 10 10
vegh landuse_cat 4 0.3 10 10 0.3
vegh landuse_cat 53 10 10 3
vegh landuse_cat 6 0.05 0.5 0.2 0.05
vegh landuse_cat 7 0.15 0.4 0.3 0.15
vegh landuse_cat 8 1.5 2.5 2.5 1.5
vegh landuse_cat 9 0.05 0.05 0.05 0.05
vegh landuse_cat 10 0.4 3 3 0.4
vegh landuse_cat 11 0.01 0.01 0.01 0.01
vegc landuse_cat 1 0.1 0.1 0.1 0.1
vegc landuse_cat 2 0.5 0.5 0.5 0.5
vegc landuse_cat 3 0.9 0.9 0.9 0.9
vegc landuse_cat 4 0.7 0.95 0.95 0.7
vegc landuse_cat 5 0.8 0.92 0.92 0.8
vegc landuse_cat 6 0.3 0.8 0.7 0.3
vegc landuse_cat 7 0.95 0.95 0.95 0.95
vegc landuse_cat 8 0.9 0.95 0.95 0.9
vegc landuse_cat 9 0.8 0.8 0.8 0.8
vegc landuse_cat 10 0.75 0.75 0.75 0.75
vegc landuse_cat 11 0.1 0.1 0.1 0.1
Table 6.7 - seasonal parameters table.

The maps generated with the h.monthly.par commands have been


derived with a time resolution of one month, while the ones
generated with the h.seasonal.par command have been extracted
with a time resolution of one week.
Higher time resolution are not justified, because the approximation
of the parameters themselves does not permit more precise
estimates.
These maps generated by means of literature values spatialization,
take in account the mean behaviour of these quantities (seasonal or
monthly general trend) but do not consider any particular

6.29
Chapter 6 – Case study: the Verzasca basin

conditions.
For instance, the LAI (Leaf Areal Index) can be greatly affected by
the presence of dry periods that would limit the vegetation growth.
In this sense remote sensed maps of LAI, vegetation height,
vegetation cover, and albedo, being an updated observation of the
reality, could be a really valuable resource for a better watershed
characterization.
The investigation of remotely sensed data is not the argument of
this thesis that indeed focus its attention to the realization of a open
source GIS embedded hydrological model, therefore further
research in this area will be remanded to future investigations.
The generated map series has been verified by visually comparing
their maps and checking their plausibility both in spatial
distribution (e.g. reflecting the landuse distribution) and in time
distribution (e.g. growing LAI values from winter to summer).
Like for climatic maps a simple check to verify the absence of
missing maps or nulls values in the generated series was
successfully executed.
In the top of figure 6.28 the albedo map for the August month is
shown. Due to its time independence, the albedo raster series shows
constant values through months; the accordance with the landuse
map (see figure 6.13) can be observed.
In order to take into account the presence of the snow, before the
running of the snow accumulation and melt module, the albedo map
will updated by means of map algebra. Where, at the previous step,
the snowpack water content was not zero the current albedo map
assume a LAI value of 3.8, value estimated to be a mean value of the
snow reflectance (Tarboton, 1996). The bottom of figure 6.28 show
an example of updated albedo map for the 15 th August in full
agreement with the estimated snow coverage (see figure 6.34).
In figure 6.29, 6.30, 6.31 two maps of the vegetation height,
vegetation cover and LAI, one for winter time and one for summer
time, are represented as an example.
In figure 6.29 the correspondence with the modified landuse map
can be noted, in fact the river cells are well represented with low

6.30
Chapter 6 – Case study: the Verzasca basin

vegetation height (~0.1 m) and the pine forest cells have larger
vegetation hight values (~10 m). Moreover the extension of red
cells (indicating higher values) became larger in summer due to the
growing of the deciduous tree and mixed forest (respectively from
~3 and ~0.3 to ~10 m).
Also in figure 6.30 the good correspondence with the landuse map
can be noted. Moreover the full blooming of the deciduous tree and
mixed forest in the summer season shows large values of coverage
(~90%) if compared to the winter season values (~0.7%), all in full
agreement with the vegetation growing cycle.
Comparing figure 6.31 with figure 6.13 the correspondence with the
landuse can be identified. The comparison between winter map and
summer map reveal, in full accordance with the evolution of the
vegetation, a general increasing LAI values from January to August.
This behaviour, motivated by the increasing leaf area of the
vegetation, is not present in the water, settlement, rock, and river
landuse classes where vegetation is mostly absent.

6.31
Chapter 6 – Case study: the Verzasca basin

Figure 6.28 – albedo map for August 2004 (top) and updated albedo map
for the 15th August 2004.

6.32
Chapter 6 – Case study: the Verzasca basin

Figure 6.29 – vegetation height [m] on the 15th January (top)


and 15th August (bottom).

6.33
Chapter 6 – Case study: the Verzasca basin

Figure 6.30 – vegetation cover [%] on the 15th January (top)


and 15th August (bottom).

6.34
Chapter 6 – Case study: the Verzasca basin

Figure 6.31 – LAI index [-] on the 15th January (top)


and 15th August (bottom).

6.35
Chapter 6 – Case study: the Verzasca basin

6.11 Net radiation


The net energy entering in the catchment system has been hourly
calculated following the procedures illustrated in figure 4.2.
The correct spatial distribution of the net radiation maps can be
detected through a comparison with the aspect map. In fact, the
side exposure of the mountains, with respect to solar position, is the
driving factor in the net radiation values establishment.
As an example, figure 6.32 shows the aspect map of the Verzasca
region and the net radiation map calculated on the 18th August
2004 at 10:00 am. At that hour of the morning the sun is positioned
at east and its radiations directly hit the side of the mountain
exposed to east. From the comparison of the two maps of Figure
6.32 the correct reproduction of this situation can be detected.
The temporal evolution of the radiation is shown in the figure 6.33,
where each frame represent the estimated net radiation ad different
hours. The presence of clouds on the weast mountains of the basin
can be observed from 11 a.m. to 4 p.m..
Validity of the estimations has been evaluated by comparing
extracted mean values with other study previously carried on in this
region (Menzel, 1997).

6.36
Chapter 6 – Case study: the Verzasca basin

Figure 6.32 – map of the aspects [degree] (top) and net radiation
[MJ/(m2·h)] for the 18th August 2004 at 10:00 am (bottom).

6.37
Chapter 6 – Case study: the Verzasca basin

Figure 6.33 – evolution of the net radiation series from 6 a.m. to 5 p.m.
On the 5th October 2004, (legend like in figure 6.32).

6.38
Chapter 6 – Case study: the Verzasca basin

6.12 Snowmelt
According to the procedures described in section 4.3, the snow
raster series (snowpack water equivalent, energy content, and
snowmelt) has been generated.
The figures 6.34-37 show some generated maps of the snow
accumulation and melting process (figure 6.34 shows the snowpack
water equivalent, figure 6.35 the snowpack energy, and figure 6.36
the snowmelt components).
From a spatial point of view the generated maps are correct, in fact
the snow is accumulated on the top of the mountains, with a
snowpack water equivalent decreasing with elevation, and the
snowmelt occurs on the limits of the snow covered areas, where
temperatures are higher.
Also the quantitative simulated values show a correct behaviour.
As an example, in figure 6.37 a growing snowpack water equivalent
in correspondence of rain events, and a snowpack energy content
growing following the day-night cycle, can be detected. Moreover,
when the energy content reaches the needed value, the snowmelt
occurs following day-night cycle, and reducing the snowpack water
equivalent and energy content.
Comparing the simulated values with the estimates carried out in
the Alpine region by Menzel (1997) a good agreement can be found.
The simulations for August 2004 show the following mean values:
 snowpack water equivalent: 281.34 mm
 snowmelt: 0.26 mm
 snowpack energy content: 31514.79 Kcal*m-2

6.39
Chapter 6 – Case study: the Verzasca basin

Figure 6.34 – snowpack water equivalent [mm] evolution (above: 17th


August 2004 at 00:00 am.; bottom:18th August 2004 at 10:00 am.)

6.40
Chapter 6 – Case study: the Verzasca basin

Figure 6.35 – snowpack energy [Kcal*m-2] evolution. Above: 17th August


2004 at 00:00 am. (night); Bottom:18th August 2004 at 10:00 am. (day).

6.41
Chapter 6 – Case study: the Verzasca basin

Figure 6.36 – snowmelt [mm] evolution (above: 17th August 2004 at


00:00 am.; bottom:18th August 2004 at 10:00 am.)

6.42
Chapter 6 – Case study: the Verzasca basin

Figure 6.37 – from top to bottom: (1) rainfall (mm), (2) snowpack water
equivalent (mm) in red and snowpack energy content in blue, and (3)
snowmelt (mm). All plots extracted for May 2004 from a sampling point.

6.43
Chapter 6 – Case study: the Verzasca basin

6.13 Evapotranspiration
The potential evapotranspiration maps, computed as described in
section 4.4, have been extracted and analysed in order to verify
their correctness.
No missing maps have been detected and the spatial behaviour
agree with consideration of landuse, exposition, and quota of the
terrain as shown in figure 6.38. The potential evaporation from a
liquid surface is in fact higher then the one from the terrain,
moreover the areas where rock is present the evapotranspiration
result lower then where the vegetation is present (landuse map in
figure 6.13).
Another important consideration, derived from the comparison of
the two maps of figure 6.38, is that the clear sky index greatly
influence the evapotranspiration values. In fact, when clouds are
present, the energy reaching the ground is lower and therefore less
energy is available for the water evaporation and for the vegetation
transpiration.

6 May 2004 h: 12.00 8 May 2004 h: 12.00

Cloudy day Sunny day

Figure 6.38 – potential evapotranspiration maps for two different days


of May (6th and 8th) at the same hour (midday) in two different sky
condition (cloudy and sunny).

6.44
Chapter 6 – Case study: the Verzasca basin

The simulations sampled in same check points, homogeneously


positioned in the region and different each other for quota,
exposure and landuse, reveal the correctness of the values from a
qualitative and quantitative point of view.
Looking at figure 6.39, where as an example the temporal evolution
of the evapotranspiration at one of the check points is plotted, can
be observed that the evaporation from a liquid surface is always
higher than from terrain and that the day-night cycle is well
reproduced.

Figure 6.39 – Potential evapotranspiration (red vertical lines) and


potential evaporation (blue lines) during May 2004.

From a quantitative point of view:


 the potential evapotranspiration in May 2004 shows a mean value
of 0.08 mm/h, a maximum of 0.92 mm/h and a minimum of 0.00
mm/h
 the potential evaporation from a liquid surface in the same month
shows a mean value of 0.14 mm/h, a maximum of 1.38 mm/h, and
a minimum of 0.00 mm/h.
All these values are in good agreement with other previous
estimates carried on in the study region (Menzel, 1997).

6.45
Chapter 6 – Case study: the Verzasca basin

6.14 Canopy interception


The canopy module running produces three raster series: the
canopy storage level, the interception drainage and the interception
loss.
Figures 6.40, 6.41 and 6.42 show the three components at 24 hour
distance (17th Aug 2004 at 12:00 and 18th Aug 2004 at 12:00).
The comparison of these figures with the figure 6.43, showing the
rain distribution generated for those times, allow us to detect the
correct spatial distribution of the phenomena. In fact the canopy
storage level is distributed where vegetation is present and the
higher values are localized where the rain is more intense (on the
17th August it is the upper part of the map).
Also the drainage from the canopy is coherent with the rain
distribution and shows non zero values where and when the canopy
is saturated (red values on figure 4.41).
To analyse the the correct behaviour of the simulated process the
generated maps has been samplet at different check points
distributed over the region. The goodness of the simulation has
been detected by plotting the temporal evolution of the sampled
values. As an example, in figure 6.44 the plot derived at one of the
sample point is represented. The correct behaviour can be observed
by the fact that:
 the canopy storage level grows with rainfall until the maximum
capacity is reached, and then decrease proportionally to the
evapotranspiration;
 the canopy drainage occurs only where rain falls and the canopy
is saturated;
 the interception losses are function of the evapotranspiration and
of the canopy water storage content;
 the interception losses, caused from evaporation of the
intercepted water, occurs wherever the canopy storage level is
not empty.
The quantitative correctness of the simulations has been verified by
means of the estimations statistical values reported in table 6.8.

6.46
Chapter 6 – Case study: the Verzasca basin

Map series mean maximum minimum


interception loss 0.02 0.4 0
interception drainage 0.38 8.99 0
canopy storage level 0.67 3.34 0
Table 6.8 – statistics in mm/h of the series generated with the
h.interception command.

These values are in agreement with previous estimated values in


the Ticino region (see Menzel, 1997).

6.47
Chapter 6 – Case study: the Verzasca basin

Figure 6.40 – canopy storage level (mm) at 24 hour distance (above 17th
Aug 2004 at 12:00, bottom 18th Aug 2004 at 12:00)

6.48
Chapter 6 – Case study: the Verzasca basin

Figure 6.41 – interception drainage (mm/h) at 24 hour distance (above


17th Aug 2004 at 12:00, bottom 18th Aug 2004 at 12:00)

6.49
Chapter 6 – Case study: the Verzasca basin

Figure 6.42 – interception losses (mm/h) at 24 hour distance (above 17th


Aug 2004 at 12:00, bottom 18th Aug 2004 at 12:00)

6.50
Chapter 6 – Case study: the Verzasca basin

Figure 6.43 – rainfall intensity (mm/h) at 24 hour distance (above 17th


Aug 2004 at 12:00, bottom 18th Aug 2004 at 12:00)

6.51
Chapter 6 – Case study: the Verzasca basin

Figure 6.44 – from top to bottom: rainfall (mm/h), evapotranspiration


(mm/h), interception drainage (mm/h), interception losses (mm/h),
canopy storage level (mm).

6.52
Chapter 6 – Case study: the Verzasca basin

6.15 Runoff calibration


As already mentioned in section 4.8, in physically based distributed
hydrological models,various parameters must be calibrated. This
can involve tremendous effort, especially when the number of
parameters is large and the applied hydrological model is
computationally expensive. Automatic parameter estimation tools
can significantly facilitate the calibration process, hence the inverse
automatic calibration model UCODE-2005 (Poeter et al., 2005) has
been combined with the developed distributed hydrological model
HydroFOSS.
Using the developed UCODE-HydroFOSS linkage, a first sensitivity
analysis was performed in order to extract information and to
identify the most important parameters to be calibrated for
reproducing a good fitting between observations and simulations.
Usually, in order to run effective calibration a period that,
depending on the basin characteristics, can vary from 3 up to 12
months is used (Montanari, 2006). This long period allows to
include all different types of events that can occurs (e.g. short
summer shower, and long winter precipitation), and to let the soil
moisture reach the real condition.
This operation can easily asks for a couple of working months
because:
• the high number of processes and calculus required for each cell
of the grid, the simulations result expensive in terms of time and
computational power;
• the calibration runs one complete running for each parameter
perturbation;
• multiple calibration process (with different adjustable parameters
with different starting values) has to be run to reach the desired
solution.
With the aim to validate the HydroFOSS model, and considering the
few time available to conclude the PhD, a first calibration has been
performed on the only month of August 2004.
Obviously this calibration can only reveals the general correctness

6.53
Chapter 6 – Case study: the Verzasca basin

of the model and the parameter values derived from this calibration
could not be valid for a good fitness in other periods.
For kindly concession of the CSCS (Swiss National Supercomputing
Center, www.cscs.ch) of Lugano, the implemented commands, the
GRASS GIS, and the UCODE-2005 program has been installed on a
HP computer with 4CPU Itanium/2 (IA-64) under a Linux (Suse
distribution) environment. Such a system allows the running of one
month simulation in approximately one and half hour.
Considering the short period considered, with the aim to reduce the
warm-up period where soil the water content is not correctly
simulated, a new parameter has been introduced: the starting soil
saturation percentage (sat).

6.15.1 First sensitivity analysis


Considering the soiltypes and the landuses present in the Verzasca
basin, the significance 36 parameters to the discherges simulation
has been analysed:
 the Manning coefficient for 11 different landuses identified by
the simbols n_X, where X represent the soiltype category index
(n_1, n_2, n_3, n_4, n_5, n_6, n_7, n_8, n_9, n_10, n_11);
 the soil conductivity for 6 different soiltypes identified by the
symbol ks_X, where X is the soiltype category index (ks_2, ks_3,
ks_4, ks_7, ks_8, ks_9);
 the saturated soil water content for 6 different soiltypes
identified by the symbol ts_X, where X is the soiltype category
index (ts_2, ts_3, ts_4, ts_7, ts_8, ts_9);
 the residual soil water content for 6 different soiltypes identified
by the symbol tr_X, where X is the soiltype category index (tr_2,
tr_3, tr_4, tr_7, tr_8, tr_9);
 the soil thickness for 6 different soiltypes identified by the
symbol L_X, where X is the soiltype category index (L_2, L_3, L_4,
L_7, L_8, L_9);
 The starting soil saturation percentage SAT.

Parameters were chosen that likely would have the greatest effect

6.54
Chapter 6 – Case study: the Verzasca basin

on channel discharges. From a physical point of view, those


parameters directly influence the water flows into the subsurface,
overland and streams.
Parameters directly responsible for the amount water available for
the runoff driving the interception drainage, the evapotranspiration
and the snowmelt were not included in the recharge-parameter
analysis. The amount of water available for the runoff was
understood to have the greatest effect on the channel flow;
therefore, those parameter sensitivities could overwhelm other
sensitivities and mask subtler effects of the runoff flow processes.
Scaled sensitivities were calculated by perturbing (increasing and
decreasing) the “a priori” parameter values of the model and
calculating the derivative of the simulated value (discharge), with
respect to the parameter. The size of the perturbation is calculated
as a user-specified factor. All parameters by 25 percent.

The importance of different parameters to the calculation of a


observed value i and the importance of different observations to the
estimation of a single parameter bj can be detected by the
dimensionless scaled sensitivity (DSSij) index.
The dimensionless scaled sensitivity, equation, is calculated as (Hill,
1998):

 ∂y ' i 
DSSij =   ⋅ b j ⋅ϖ i1/ 2
 ∂b  [eq. 6.2]
 j

where:
i identifies one of the observations,
j identifies one of the parameters,
y'i is the simulated value associated with the i-th observation(in
this case it is the hourly measured discharge at the two
limnigraph of figure 6.19),
bj is the j-th estimated parameter (in this case they are the 36
analysed parameters),
∂y'i/∂bj is the sensitivity of the simulated value associated with the i-

6.55
Chapter 6 – Case study: the Verzasca basin

th observation and is evaluated at the final parameter values,


and
ϖi is the weight for the i-th observation (in this case it is
assigned equally to each observation).

In figures 6.45 and 6.46 the obtained DSS relative to station 1 and
station 2 has been respectively plotted.
Keeping in mind that for one parameter the observation in
correspondence of higher DSS have higher importance for its
estimation, and that for one observation the parameter with higher
DSS has higher importance in the simulation of the correspondent
simulation, these figure can be analysed and same consideration
can be done.
The figures show that in the first half month (from 0 th to ~330th
hours) the starting saturation percentage (SAT) is the most
important parameter for the best fitting of the simulation, while the
Manning coefficient for the river landuse (n_11) results in being
quite important for all over the period.
Other sensitive parameters are the soil conductivity and the soil
thickness for the regosol eutric predominant soiltype (ks_2 and
L_2), which is the most diffuse soiltype in the catchment.

6.56
Dimensionless Scaled Densitivity (station 1)

20 160

SAT
15 140
n_1
n_2
n_3
n_4
n_5
10 120 n_6
n_7
n_8
n_9
n_10
L_2 n_11
5 100 ks_2
ks_7
ks_8
ks_2 ks_9
Ts_2
Ts_3
0 Ts_4
80

DDS
Ts_7

6.57
Ts_8
Ts_9
Tr_2
discharges [m3/s]

Tr_3
Tr_4
-5 60 Tr_7
Tr_8
Tr_9
L_2
n_11 L_3
L_4
-10 40 L_7
L_8
L_9
SAT
discharges

-15 20

observed discharges
-20 0

Figure 6.45 – Dimensionless Scaled Sensitivity of the 36 model


1 25 49 73 97 121 145 169 193 217 241 265 289 313 337 361 385 409 433 457 481 505 529 553 577 601 625 649 673 697

parameters, and the relative observed river discharge at station 1.


hours
Chapter 6 – Case study: the Verzasca basin
Dimensionless Scaled Densitivity (station 2)

20 500

450
15 n_1
n_2
n_3
SAT 400 n_4
n_5
n_6
n_7
10 n_8
350 n_9
n_10
n_11
ks_2
ks_7
300 ks_8
5 ks_9
Ts_2
ks_2 Ts_3
Ts_4
250

DSS
Ts_7

6.58
L_2 Ts_8
Ts_9
0 Tr_2
discharges [m3/s]

200 Tr_3
Tr_4
Tr_7
Tr_8
Tr_9
n_11 150 L_2
-5 L_3
L_4
L_7
L_8
100 L_9
SA T
discharges
-10
50
observed discharges
-15 0

Figure 6.46 – Dimensionless Scaled Sensitivity of the 36 model


1 25 49 73 97 121 145 169 193 217 241 265 289 313 337 361 385 409 433 457 481 505 529 553 577 601 625 649 673 697

parameters, and the relative observed river discharge at station 2.


hours

A better view of the parameters influence can be achieved by


Chapter 6 – Case study: the Verzasca basin
Chapter 6 – Case study: the Verzasca basin

plotting the maximum DSS for each observation with information on


the parameter type, as shown in figure 6.47: this plot highlight
which parameter is the most important for the correctness of the
fitting along the time axis.
Four different periods of influence can be detected:
• a first period of warm-up where SAT is driving the fitting (0 –
330);
• two periods of low flow where the n_11 is driving the fitting (330-
430 and 480-720);
• one period of rising flow where L_2 id driving the fitting (430 –
450);
• one period of recession flow where ks_2 is driving the fitting (450
– 480).
These influence periods can be explained as follows:
• the model need of a first period to well represent the real soil
saturation level, in this phase the starting saturation level
strongly affect it;
• the rising limb of the hydrograph is mainly due to the overland
flow (fast flow component) that occur only when the soil is
saturated: the soil thickness influences the soil water capacity
and therefore regulates the water excess (amount of water
available for overland flows);
• the recession limb of the hydrograph is mainly due to the
subsurface flow (slow flow component) whose velocity, and
therefore the amount of water that reach the streams at each
hour, is driven from the soil conductivity parameter;
• in low flow condition the amount of water in the river and its
velocity is driven from the roughness coefficient of Manning.

6.59
80 160
SAT n11 L2 ks2 n11

140

60

120

40

100

LA RGE S T D S S
20 80 P A R_type
discharges

6.60
60

40

-20

20

Figure 6.47 – largest calculated DSS for station 1


-40 0
1 25 49 73 97 121 145 169 193 217 241 265 289 313 337 361 385 409 433 457 481 505 529 553 577 601 625 649 673 697
Chapter 6 – Case study: the Verzasca basin
Chapter 6 – Case study: the Verzasca basin

The total amount of information provided by the observations for


the estimation of one parameter bj can be calculated by means of
the Composite Scaled Sensitivity (CSS) index.
CSS summarize all the sensitivities for one parameter. CSS are
calculated for each parameter using the dimensionless scaled
sensitivities for all observations. Because they are dimensionless,
CSS can be used to compare the amount of information provided by
different types of parameters. Model simulation results will be more
sensitive to parameters with large CSS relative to those for other
parameters. The CSS for the j-th parameter, CSSij , is calculated as
(Hill, 1998):

ND

[[  DSS  ] / ND ]
1/2
CSS j =∑ [eq. 6.3]
2
ij b
i=1

where:
ND is the number of observations (discharges measured at
station 1 and 2) being used in the regression fitting,
bi is a vector that contains the parameter values at which the
sensitivities are evaluated (the 36 parameters), and
DSSij equals the scaled sensitivities of equation 6.2.

Figure 6.48 showing a plot of the calculated CSS for the different
parameters, highlights, the overall importance of the Manning
coefficient of the river (n_11), of the conductivity of the regosol
eutric predominant soiltype (ks_2), of the soil thickness and of the
saturated soil water content of the most diffuse soiltypes (L_2, L_3,
L_4, ts_2, ts_3, ts_4). Of course the starting saturation percentage of
the soil results in being the most important due to the already
mentioned warm-up period, when it drives the goodness of the
fitting.

6.61
Chapter 6 – Case study: the Verzasca basin

Composite Scaled Sensitivity

1.60E+01

1.40E+01

1.20E+01

1.00E+01

8.00E+00

6.00E+00

4.00E+00

2.00E+00

0.00E+00

Ts_2
Ts_3
Ts_4

Ts_8
Ts_9
Tr_2
Tr_3
Tr_4
Tr_7

SAT
Ts_7

Tr_8
Tr_9
n_1
n_2
n_3
n_4
n_5
n_6
n_7
n_8
n_9
n_10
n_11
ks_2
ks_3
ks_4
ks_7
ks_8
ks_9

L_2
L_3
L_4
L_7
L_8
L_9
Figure 6.48 – Composite Scaled Sensitivity of the 36 parameters
analysed.

The influence of a given observation on the simulation fitting due to


its location in the space of the inputs can be evaluated by means of
the LEVERAGE statistics. Looking at figure 6.49 where the
LEVERAGE calculated for the two station, and the discharges for
station 2 are plotted, can be noted that, not taking in account the
first warm-up period, the most important observations for a good
fitness are the one positioned at the beginning of the first rising
limb of an event. Thus those observation should have a large weight
in order to force the simulation to closely estimates them and
therefore make the model having a good fitting with observations.

6.62
Chapter 6 – Case study: the Verzasca basin

---- Leverage of station 1


---- Leverage of station 1
----LEVERAGE
observed discharges of station 2
1.00E+20 500

9.00E+19 450

8.00E+19 400

7.00E+19 350

6.00E+19 300
leverege

leverege (station 1)
5.00E+19 250 leverege (station 2)
discharges st 2

4.00E+19 200

3.00E+19 150

2.00E+19 100

1.00E+19 50

0.00E+00 0
1 28 55 82 109 136 163 190 217 244 271 298 325 352 379 406 433 460 487 514 541 568 595 622 649 676 703
hours

Figure 6.49 – LEVERAGE statistics for station 1 and 2.

6.15.2 First model calibration


The calibration was run accordingly to section 4.8.3 by means of the
UCODE calibration inverse model. Inverse modelling, or parameter
estimation modelling, is an automated calibration technique that
works by finding parameter values (e.g., hydraulic conductivities,
roughness coefficients) that minimize the sum of the squares errors,
also called the objective function, for a given model configuration.
As a result of the sensitivity analysis carried out in the previous
section (6.14.1), it has been decided to focus the calibration process
upon 8 parameters: n_11, ks_2, ks_3, ks_4, L_2, L_3, L_4, and SAT.
The starting values adopted during the calibration (shows in table
6.9) where chosen as found in literature (Liu and Todini, 2002),
(Foglia at al., 2006) where the TOPKAPI model was applied.
In order to reduce the iterations, only the most important parameter
of ks and L set was directly calibrated. The others parameters were

6.63
Chapter 6 – Case study: the Verzasca basin

proportionally modified at each iteration following the equation:

Pi = Pj * Pj'/Pi' [eq. 6.4]

where Pi is the parameter i, Pj the reference parameter j, Pj' and Pi'


are the starting values of Pj and Pi at the beginning of the
calibration.
Therefore the introduced relations were:
ks_3 = ks_2*0.000592/0.000654
ks_4 = ks_2*0.000592/0.000620
L_3 = L_2*0.2/0.6 [eq. 6.5]
L_4 = L_2*0.2/0.2

in this way the calibration were actually performed only on 4


parameters: n_11, ks_2, L_2, and SAT.
The physical meaning of the results has been controlled by applying
upper and lower constrains to the starting values and maximum
fractional parameter change allowed between parameter iteration.

n_1 0.210000 Ts_2 0.680000


n_2 0.100000 Ts_3 0.600000
n_3 0.350000 Ts_4 0.520000
n_4 0.250000 Ts_7 0.525000
n_5 0.250000 Ts_8 0.400000
n_6 0.250000 Ts_9 0.450000
n_7 0.300000 Tr_2 0.080000
n_8 0.400000 Tr_3 0.120000
n_9 0.150000 Tr_4 0.120000
n_10 0.150000 Tr_7 0.140000
n_11 0.040000 Tr_8 0.020000
ks_2 0.000600 Tr_9 0.125000
ks_7 0.000055 L_2 0.400
ks_8 0.000001 L_7 0.250
ks_9 0.000130 L_8 0.375
SAT 40.00000 L_9 0.080

Table 6.9 – starting values for the calibration: in coloured cells the
calibrated parameters.

6.64
Chapter 6 – Case study: the Verzasca basin

The calibration result shows the following values:


- n_11 (river Manning coefficient) = 2.3E-02 [s/m1/3];
- ks_2 (regosol soil conductivity) = 0.678E-03 [m/s];
- L_2 (regosol soil thickness) = 0.374 [m];
- SAT (starting soil saturation) = 7.6 [%].

Consequentially the other scaled values are:


- ks_3 = 0.614E-03 m/s
- ks_4 = 0.647E-03 m/s
- L_3 = 0.125 m
- L_4 = 0.374 m

The calibrated values have physical meaning. A partially low value


of the river Manning coefficient could be explained with the
presence of a rocky river bed for a long part of the Verzasca.
The simulation generated with the calibrated parameters has been
plotted against the observations for the two different stations in
figures 6.50 and 6.51.

6.65
station 1 - calibration

160

140

120

100

observed
80
simulated

6.66
discharges [m3/s]
60

discharges at station 1.
40

20

Figure 6.50 – Observed and post calibration simulated


0
1 25 49 73 97 121 145 169 193 217 241 265 289 313 337 361 385 409 433 457 481 505 529 553 577 601 625 649 673 697
hours
Chapter 6 – Case study: the Verzasca basin
station 2 - calibration

500

450

400

350

300

observed
250

6.67
simulated

discharges [m3/s]
200

discharges at station 2.
150

100

50

Figure 6.51 – Observed and post calibration simulated


0
1 25 49 73 97 121 145 169 193 217 241 265 289 313 337 361 385 409 433 457 481 505 529 553 577 601 625 649 673 697
hours
Chapter 6 – Case study: the Verzasca basin
Chapter 6 – Case study: the Verzasca basin

Looking at the figures 6.50 and 6.51 the hydrograph behaviour is


well represented, and large peaks occurred at the 457 th hour appear
to be well fitted. This is particularly important, in fact, in floods
alert application, be able to reproduce high discharge can help in
save life and properties.
The quite good low flows fitting could be partially explained by the
nature of the basin, that does not present significant base flows
(process not simulated in the model), and by the presuming well
simulated snowmelt process.
Calibration improvement are needed to better simulate the
recession limb of the hydrograph.
The total error produced by the model simulation, by means of the
total discharge in August, is the 37%.
This values, has to be considered satisfactory because this is just
the result of a first calibration study on a short period and because
the large uncertainties in distributed hydrological modelling hardly
lead to much better results.

The validity of the runoff modules can be detected looking at figures


6.52, 6.53, 6.54, 6.55 where the channel water discharge, the
overland water discharge, the soil water discharge, and the soil
water content are represented before, during, and after the intense
rain event of the 20th August as an example.
In the figure 6.53 not zero values are present only where the is the
river, the values grown accordingly with flow direction, and they
rise and decrease following the rain event evolution (from A to C).
Therefore the validity of the spatial distribution and behaviour is
assured.
The overland flows are generally not present when there is not an
heavy or persistent rain (the saturation excess is null), then when
the event occurs the soil get saturated and the overland discharges
start. The first cell where the overland flow occurs are the ones with
higher flow accumulation values, in fact they are saturate from the
water directly coming from the rain and from the flows (overland
and subsurface) coming from the up-land cells. The figure 6.54 well

6.68
Chapter 6 – Case study: the Verzasca basin

represent the spatial distribution of the overland discharge and its


temporal behaviour.
The flows in the soil is slow (subsurface component) is slower then
the other two, and the decreasing of the discharges need longer
time to be significant. Moreover the magnitude of the discharge
depends on the hydraulic head, thus soiltypes with lower thickness
can intake less water and will produce lower discharge values, while
cells at the bottom of the valley, that receive water from all the
upland ones, will produce higher discharges. In figure 6.55 this
spatial distribution and temporal behaviour can be found. The low
subsurface discharge values in correspondence of the river are
because those the cells are not fully covered by soil, resulting in
reduced discharges.
The soil water content rise until a maximum capacity, that depends
on the soiltype, is reached. As already mentioned the filling of the
soil can be done by direct rainfall or upland discharges, thus higher
values can be observed at the bottom of the valley or where an
heavy rain is present. This component is emptied by
evapotranspiration and subsurface flows, thus the water content
decrease slowly. Figure 6.56 show the described spatial and
temporal behaviour. Fully distributed colors in the map, represent
saturated area.
In all these maps, the absence of heavy rain, in the lower east part
of the basin, is correctly represented in all the components. Here
the channel flow does not increase, no overland flow occurs, low
soil water discharges are simulated, and the soil water content do
not increase.

6.69
Chapter 6 – Case study: the Verzasca basin

A B

Figure 6.52 – channel water discharge maps [m3/s].


A: before an event (17th August 2004, 16:00).
B: during the peak (20th August 2004, 06:00).
C: during the recession (20th August 2004, 16:00).

6.70
Chapter 6 – Case study: the Verzasca basin

A B

Figure 6.53 – overland water discharge maps [m3/s].


A: before an event (17th August 2004, 16:00).
B: during the peak (20th August 2004, 06:00).
C: during the recession (20th August 2004, 16:00).

6.71
Chapter 6 – Case study: the Verzasca basin

A B

Figure 6.54 – soil water discharge maps [m3/s].


A: before an event (17th August 2004, 16:00).
B: during the peak (20th August 2004, 06:00).
C: during the recession (20th August 2004, 16:00).

6.72
Chapter 6 – Case study: the Verzasca basin

A B

Figure 6.55 – soil water content maps [m3].


A: before an event (17th August 2004, 16:00).
B: during the peak (20th August 2004, 06:00).
C: during the recession (20th August 2004, 16:00).

6.73
Chapter 7 – Conclusions and future works

7. Conclusions and future works


Nowadays Geographical Information Systems are projected in the
era of modelling in order to allow a conscientious environmental
management of the resources. In agreement with this tendency, in
this work the GIS capability to model the rainfall-runoff has been
proven by means of the development of procedures and new
commands simulating all the physical processes driving the water
flows (HydroFOSS model). Distributed hydrological modelling,
allows us to predict the impact of landuse changes and climatic
change on water resource and on hydrological regimes of river
basins.

Considering the natural capability of GISs to handle, store,


represent and analyse spatial information, the embedding of such a
kind of model into a GIS system establishes a valuable resource for
solving numerous management problems (flood mitigation, storm
drainage, bridges, sewerage and culverts design, water supply,
irrigation, hydro power, navigation, pollution control, fresh water
conservation, etc.).
The usage of a Free and Open Source (FOSS) GIS Software, largely
facilitates the task of new procedures and model development due
to the accessibility of its code, and availability of documentation.
Moreover, a GIS-embedded approach leads to better accuracy, less
duplication, easier map storage, more flexibility, more data sharing
and interoperability, timeliness, more efficiency, higher product
complexity, and full control of the spatial component.
The followed modular approach, in conjunction with the standard
GIS commands and the accessibility of the code, guarantees, to an
expert user, the fully customization of the application.

During this work a new approach for raster time series data
management, and a database to archive in real time observations of
three meteorological station networks, have been developed.

7.1
Chapter 7 – Conclusions and future works

This research has also pointed out the need of a new vector library
design able to handle time series with a constant spatial component
and a variable attribute component dynamically extracted from a
database.

A procedure for input map generation has been developed.


The input climatic raster maps has been generally derived by means
of Thiessen tessellation procedure; this “rough” techniques of
interpolation is commonly adopted in hydrological modelling due to
the lack of sufficient spatially distributed observation.
In order to improve the temperature field estimation, different
methods were investigated by means of minimum MAE (mean
absolute error). For the Canton Ticino region the linear gradient
approach resulted the more appropriate technique.
Input maps of parameters or index characterizing the watershed are
also required from distributed hydrological models; these maps has
been spatialized by means of soil-landuse characteristics, current
time of the year, and altitude.
Climatic and watershed characterizing maps, could be derived from
remotely sensed data, although their validity, in term of magnitude
of the estimates, has to be proven.
It was carried on a first study of the rainfall estimations derived
from ground meteorological radar observations. The results of this
study showed that the spatial information is correctly estimated,
while the quantitative information is not jet accurate.
More researches in this area, and some application test will be
argument of future works.

New commands simulating the snow accumulation and melting,


potential evapotranspiration, canopy interception processes, and
new procedure to automatically calculate the net radiation have
been developed. A first validation has proven their validity, in terms
of spatial representation of the phenomena, and mean quantitative
values.

7.2
Chapter 7 – Conclusions and future works

A new command for the simulation of the runoff process has been
generated and included in a main shell script, that automatically
calls, in the appropriate order, all the required procedures and
commands to simulate the entire rainfall-runoff process.
Although the model is entirely physically based, its parameters
(with a physical meaning), are not directly measurable on the field.
In fact they are mean values referred to a cell area (100 m 2 in the
case study) where high heterogeneities can be found.
This uncertainties in parameters values ask for calibration
procedures aiming at estimate their correct values.
For this reason a linkage between the new developed HydroFOSS
model and the FOSS program UCODE-2005 for automatic inverse
calibration has been established.

The validity of the runoff module has been proven by means of a


sensitivity analysis and calibration in the Verzasca basin. The
generated maps are in full accord with all the expected temporal
and spatial behaviour of the three runoff components.

Due to the lack of time and the high computational time and power
requested for such a kind of model, the sensitivity analysis and the
calibration was run on a short period of one month.
Beneath this limited period, the sensitivity analysis pointed out the
importance of four different parameters:
• the starting soil saturation level, driving the warm-up period
needed to represent the correct soil saturation state. Calibrating
this value can result in reduced warm-up periods required.
• The soil conductivity, drives the drainage of the subsurface soil
component which mainly characterize the recession limbs of the
hydrograph. Thus its correct calibrated value allows a good fit of
the simulations in recession periods.
• The soil thickness, drives the saturation excess trigger. The water
exceeded quickly flows through the overland component to the
river, where a flood peak is generated. Therefore this parameter
regulates the goodness of the fit during the rising discharges

7.3
Chapter 7 – Conclusions and future works

periods.
• The river Manning coefficient, being a roughness coefficient, has
a friction function driving the river flow velocity. Thus it is
important for the correct simulation of dry periods, where the
available channel water is quite constant.

The calibration was carried out for the month of August 2004 by
using the hourly observations of two limnigraph stations and reveals
satisfactory results. The hydrograph behaviour is well represented
and the pecks are well fitted.

Sensitivity analysis and calibration depend on the starting


parameter values and on the observation values. Therefore different
observed periods or catchments, could lead to different calibrated
values and model responses. In order to extend the validity of the
simulation through time and space more calibrations are needed.

Three main future activities of research have been detected:


• Sensitivity analysis and calibration, on longer periods and
different basins. In this way different rainfall event types
(thunderstorm, shower, drizzle, etc.) and different field
conditions (slopes, aspects, landuses, soiltypes, etc.) can be
considered.
• Single module validation. Although the model calibration
indirectly accomplishes all the others developed modules, and a
first validation of each one has been carried out, a validation
executed with direct measurements (not available at this time)
will greatly improve the validity of each modules.
• Remotely sensed map investigation. The ability of this technique
to well describe the spatial distribution of the phenomena could
result in a big improvement for distributed hydrological
modelling once quantitatively verified. In particular, for the
rainfall intensity, it could be interesting to evaluate the
importance of the gain in spatial information (derived from the
use of radar data) versus the gain in magnitude measurements

7.4
Chapter 7 – Conclusions and future works

(derived from the use of rain gauges) in hydrological modelling.

7.5
Acknowledgment

Acknowledgment
This research has been possible due to the support of the IST
(Istituto Scienze della Terra, Institute of Earth Sciences), belonging
to the DACD (Dipartimento di Architettura, Costruzioni e Design,
Department of Architecture, Constructions and Design) of the
SUPSI (Scuola Universitaria Professionale della Svizzera Italiana,
University School of Applied Sciences of the Southern Switzerland).

A special thanks to Prof. F. Sansò, Prof. M.A. Brovelli and to all the
people working at the “Laboratorio di Geomatica” (Geomatic
Laboratory) of the “Como Campus” of the “Polytechnic of Milan”
that help me in the project development.

Thanks to Helena Mitasova, from the Dept. of Marine, Earth and


Atm. Sciences of the North Carolina State University in Raleigh
(http://skagit.meas.ncsu.edu/~helena/): here notes and suggestions
help me in correcting this document and making its contest better.

Last but not least, a big thanks to all my family and people I love
that allows me to get this PhD degree.

8.1
References

REFERENCES

Abbott, M. et al., 1986a, An introduction to the European


Hydrological System – Système Hydrologique Européen,
SHE; 1. History and philosophy of a physically based
distributed modeling system, Journal of Hydrology, Vol. 87,
45-59.
Abbott, M. et al., 1986b, Physically based distributed modelling of
an uplamd catchment using the Système Hydrologique
Européen, Journal of Hydrology, Vol. 87, 79-102.
Allen, R.G., L.S. Pereira, D. Raes, and M. Smith. 1998. Crop
Evapotranspiration: Guidelines for computing crop water
requirements. Irrigation and Drainage Paper 56, Food and
Agriculture Organization of the United Nations, Rome, pp.
300
Antenucci, J. C., Brown, K., Croswell, P. L. and Kevany, M. J., 1991.
Geographic Information Systems: A Guide to the
Technology.
Beasley, D. B., Huggins, L.F., and Monke, E.J. 1980. ANSWERS: A
model for watershed planning. Trans. of the ASAE
23(4):938-944.
Beck, M. B., 1991. Forecasting environmental change, J. Forecast.,
10, 3–19.
Beven, K. J., 1981, Kinematic subsurface stormflow, Water
Resources Research, 17(5): 1419-1424.
Beven, K. J., 1982, On subsurface stormflow: prediction with
simple kinematic theory for saturated and unsaturated
flows, Water Resources Research, 18(6): 1627-1633.
Beven, K. J., 1989, Changing ideas in hydrology: the case of
physically-based models, Journal of Hydrology, 105: 157-
172.
Beven, K. J., 1997, TOPMODEL: a critique, Hydrological Processes,
vol. 11, 1069-1085.

9.1
References

Beven, K. J., 2001, Rainfall-runoff modeling – The Primer.


Chichester, John Wiley & Sons Ltd, 360
Beven, K. J., and Kirkby, M.J., 1979, A physically based, variable
contributing area model of basin hydrology, Hydrological
Science Bulletin, 24(1): 43-69.
Beven, K.J., Lamb, R., Quinn, P., Romanowicz, R., and Freer, J.,
1995, TOPMODEL, in Computer Models of Watershed
Hydrology, Singh V.P. (ed.), Water Resources Publications,
627-668.
Bian, L., Sun, H., Blodgett, C., Egbert, S., Li,W., Ran, L., and
Koussis, A., 1996. An integrated interface system to couple
the SWAT model and ARC/INFO. Proc., 3rd Int.
Conf./Workshop on Integrated Geographic Information Sys.
and Envir. Modeling., UC Santa Barbara.
Blake, G.J., 1975. The interception process. In: Prediction in
catchment hydrology. Australian Academy of Science,
Canberra, Australia, pp. 59-81.
Borga M., Tonelli F., Moore R.J. and Andrieu H., 2002. Long-term
assessment of bias adjustement in radar rainfall estimation,
Water Resour. Res., Vol. 38, No. 11, pp. 1226-1235.
Chorley, R. J., and Haggett, P., 1967. Models in geography.
Methuen, London.
Chow, V .T., 1973. Open Channel Hydraulics, McGraw-Hill, New
York.
Chumchean S, Sharma A. and Seed A., 2003. Radar rainfall error
variance and its impact on radar rainfall calibration, Phys.
Chem. of the Earth, Vol.28, pp. 27-39.
Ciarapica, L., Todini, E., 2002. TOPKAPI: a model for the
representation of the rainfall-runoff process at different
scales, Hydrological Processes 16, 207-29.

9.2
References

Collins F.C.Jr., Bolstad, P.V., 1996. A Comparison of Spatial


Interpolation Techniques in Temperature Estimation.
Proceedings of the Third International
Conference/Workshop on Integrating GIS and
Environmental Modeling. January 21-25, 1996, Santa Fe,
New Mexico, USA
COST Project 72, 1985. European Commission Report EUR 10353.
Crain, I.K. And MacDonald, C.L. ,1984. From land inventory to land
managment – the evolution of an operational GIS.
Chartographica 21: 39-46.
Crawford, N.H. and Linsley, R.K., 1966. Digital simulation in
Hydrology: Stanford Watershed Model IV, Technical Report
n. 39, Stanford University, Palo Alto, California.
Dawdy, D.R. and O’Donnel, T., 1965. Mathematical models of
catchment behavior, J. Hydraul. Div., Am. Soc. Civ. Eng., 91
(HY4), 123-127.
De Roo, A.P.J., Wesseling, C.G., Van Deursen, W.P.A., 2000.
Physically-based river basin modelling within a GIS: The
LISFLOOD model. Hydrological Processes 14, 1981-92.
DeBarry, P. A., 1991. GIS: Applications in nonpoint source
pollution assessment. Proc., ASCE 1991 Nat. Conf. on Hydr.
Engrg., ASCE, Reston, Va.
Doe, W. W., Saghafian, B., and Julien, P. Y., 1995. Land-use impact
on watershed response: The integration of two-dimensional
hydrologic modeling and GIS. Hydrological Processes, 10,
1503–1511.
Dunne, T. and Black, R.D., 1970. Partial area contributions to
storm runoff in a small New England watershed, Water
Resources Research, 6(5), 1296-1311.
Dunne, T., 1978. Field studies of hillslope flow process. In M.J.
Kirkby (Ed.), Hillslope Hydrology, Wiley, New York, 227-
293.

9.3
References

Dunnie, T., Moore, T.R., Taylor, C.H., 1978. Recognition and


prediction of runoff-producing zones in humid regions.,
Hydrological Sciences Bullettin 20(3): pp. 305-327.
Fetter, C.W., 1994. Applied Hydrogeology,Third Edition, Chapter
10, Macmillan.
Foglia, L., Mehl, S.W., Hill, M.C., Perona, and Burlando, P., 2006.
Evaluation of information criteria, linear statistics, cross
validation, and predictions using alternative groundwater
models, submitted to Groundwater.
Freeze, R.A. and Harlan, R.L., 1969, Blueprint for a physically-
based digitally simulated hydrologic response model, Journal
of Hydrology, n. 9, 237-258.
Freeze, R.A. and Harlan, R.L., 1969, Blueprint for a physically-
based digitally simulated hydrologic response model, Journal
of Hydrology, n. 9, 237-258.
Gjersten U., Salek M. and D.B. Michelson, 2003. Gauge-adjustment
of radar-based precipitation estimates – a review. Cost
Action 717, Working Document WDD_02_200310_1.
GRASS Development Team, 2006. GRASS 6.0 Users Manual. ITC-
irst, Trento, Italy. Electronic document:
http://grass.itc.it/grass61/manuals/html61_user/
Green, W.H. and Ampt, C.A., 1911. Studies on soil physics: 1. Flow
of water and air through soils. Journal of agricultural
Science, 4, 1-24.
Hamlett, J. M., Miller, D. A., Day, R. L., Petersen, G. W., Baumer,
G. M., and Russo, J., 1992. Statewide GIS-based ranking of
watersheds for agricultural pollution prevention. J. Soil and
Water Conservation, 47, 300–404.
Harr, R.D., 1981. Some characteristics and consequences of
snowmelt during rainfall in western oregon, Journal o f
Hydrology, nr. 53, pp. 277-304.
Harris, T. M. and Elmes, G. A., 1993. The Application of GIS in
Urban and Regional Planning: a review of the North
American experience. Applied Geography 13(1), 9-27.

9.4
References

Henderson, F.M., Wooding, R.A., 1964. Overland flow and


groundwater flow from a steady rainfall of finite duration, J.
Geophys. Res., 69 (6), 1531-1540.
Hill M.C., 1998. Methods and guidelines for effective model
calibration. US Geological Survey; Water resource
investigation report, 98-4005.
Hill, M.C. and C.R. Tiedeman, 2005. Effective calibration of
environmental models, with analysis of data, sensitivities,
predictions, and uncertainty, in press.
Hill, M.C., Banta, E.R., Harbaugh, A.W., and Anderman, E.R., 2000.
MODFLOW-2000, The U.S. Geological Survey modular
ground-water model, User’s guide to the observation,
sensitivity, and parameter-estimation processes: U.S.
Geological Survey Open-File Report 00-184, 209 p.
History of ESRI. Accessed 01/17/06:
http://www.esri.com/company/about/history.html
Hofierka, J., Suri, M., 2002. The solar radiation model for Open
source GIS: implementation and applications. Manuscript
submitted to the International GRASS users conference in
Trento, Italy, September 2002.
Horton, R.E., 1933, The role of infiltration in the hydrologc cycle.
Transactions, American Geophysical Union, 14:446-460.
INTERA TYDAC Technologies Inc. 1992. SPANS Reference
Manual. Ottawa.
Joss J., Schädler B, Galli G., Cavalli R., Boscacci M., Held E., Della
Bruna G., Kappenberger G., Nespor V. and Spiess R., 1997.
Operational use of radar for precipitation measurements in
Switzerland, www.meteoswiss.ch.
Kasten F., 1996. The Linke turbidity factor based on improved
values of the integral Rayleigh optical thickness. Solar
Energy, 56 (3), 239-244.

9.5
References

Knisel, W. G. 1993. GLEAMS: Groundwater loading effects of


agricultural management systems, University of Georgia,
Coastal Plains Experiment Station, Biological and
Agricultural Engineering Department, Publication No. 5, pp.
260
Knyazikhin, Y., Glassy, J., Privette, J.L., Tian, Y., Lotsch, A., Zhang,
Y., Wang, Y., Morisette, J.T., Votava, P., Myneni, R.B.,
Nemani, R.R., Running, S.W., 1999. MODIS Leaf Area
Index (LAI) and Fraction of Photosynthetically Active
Radiation Absorbed by Vegetation (FPAR) Product (MOD15)
Algorithm Theoretical Basis Document,
http://eospso.gsfc.nasa.gov/atbd/modistables.html, 1999.
Krajewski W.F., 1987. “Co-kriging radar-rainfall and rain gauge
data”, J. Geophys. Res., Vol. 92 , pp. 9571-9580.
Liu, Z, and Todini, E., 2002, Towards a comprehensive physically-
based rainfall-runoff model, Hydrology and Hearth System
Science, 6, 859-881.
Liu, Z., Martina, M. L.V., and Todini, E., 2005, Flood forecasting
using a fully distributed model: application of the TOPKAPI
model to the Upper Xixian catchment, Hydrology and
Hearth System Sciences, 9(4), 347-364.
Lombardo F. and Staggi L., 1998. Verifica e taratura dinamica
della strumentazione pluviometrica finalizzata alla
valutazione degli errori per intensità di pioggia elevata,
Proc. XXVI Conv. Idr. Costr. Idrauliche, Genova, Vol. 2, pp.
85-96.
Madsen, H., 2000. Automatic calibration of a conceptual rainfall–
runoff model using multiple objectives, Journal of
Hydrology, 235 (2000), 276–288.
Madsen, H., 2003. Parameter estimation in distributed hydrological
catchment modeling using automatic calibration with
multiple objectives, Advances in water resources, 26, p. 205-
216.

9.6
References

Maidment, D. R., 1991. GIS and hydrologic modeling. Proc., 1st


Int. Symp./Workshop on GIS and Envir. Modeling.
Mark, D. M., Chrisman, N., Frank, A. U., McHaffie, P. H. and
Pickles, J., 1997. The GIS History Project. Retrieved 15
January 2001 from the World Wide Web:
http://www.geog.buffalo.edu/ncgia/gishist/bar_harbor.html
Menzel L. Modellierung der Evapotranspiration im System Boden-
Pflanze-Atmosphäre, 1997. PhD Dissertation, Federal
Institute of Technology Zurich, Geographical Institute Book
No. 67 (in german).
Menzel, L., 1997. Modelling canopy resistances and transpiration
of grassland. Physics and Chemistry of the Earth, special
issue: Atmospheric and Hydrological Processes and Models
at the Soil Vegetation Atmosphere Interface.
Meteorological Organization (WMO), 1981. Technical Note N° 172,
WMO-No. 554, Geneva, Switzerland, pp. 121-123.
Michelson D.B., 2003. Quality control of weather radar data for
quantitative application, PhD thesis, Telford Inst. Env.
System., Salford University, 281 pp.
Mitas, L. and Mitasova, H., 1999. Spatial Interpolation. In:
P.Longley, M.F. Goodchild, D.J. Maguire, D.W.Rhind (Eds.),
Geographical Information Systems: Principles, Techniques,
Management and Applications, Wiley.
Mitasova, H. and Mitas L., 1993. Interpolation by Regularized
Spline with Tension: I. Theory and implementation.
Mathematical Geology, no. 25, pp. 641-655.
Mitasova, H. and Mitas, L., 2001, Modeling Physical Systems, In:
Geographic Information Systems and Environmental
Modeling, Parks B., Crane M. and Clarke, K eds., Prentice
Hall, 189-210.

9.7
References

Montanari A. and E. Toth, 2006. Calibration of spatially-distributed


hydrological models: possible strategies and analysis of the
effect of data availability. Presentation at the EGU General
Assembly, Vienna, Austria, 02-07 April 2006.
(www.costruzioni-idrauliche.ing.unibo.it/people/alberto,
2006).
Morris, E.M., 1991. Physics-based models of snow. In: Bowles,
D.S., O’Connel, P.E. (Eds.), Recent Advances in Modeling of
Hydrological Systems, Kluwer Academic, Dordrecht, pp. 85–
112.
Neitsch, S.L., Arnold, J.G., Kiniry, J.R., Williams, J.R., King, K.W.,
2002, Soil and Water Assessment Tool Theoretical
Documentation, version 2000, Texas Water Resources
Institute, College Station, Texas, TWRI Report TR-191.
Neteler M., 2005. GRASS 6 Programmer's Manual, GRASS
Development Team, Published under GNU Free
Documentation License (GFDL),
http://mpa.itc.it/markus/grass61progman/
Nizeyimana, E., 1997. Quantification of NPS pollution loads within
Pennsylvania. Publ. No. ER97-08, Envir. Resour. Res. Inst.,
Pennsylvania State University, University Park, Pa.
Ogden, F. L. 1998 . CASC2D version 1.18 reference manual. Dept.
of Civil & Environmental Engineering Rep. U-37, CT1665-
1679, Univ. Connecticut, Storrs, Conn.
Pandey, A. and Sahu, A.K., 2002, Generation of curve number
using Remote Sensing and Geographic Information System,
GIS Development,
http://www.gisdevelopment.net/application/nrm/water/water
shed/watws0015.htm
Penman, H. L. 1948. Natural evaporation from open water, bare
soil and grass. Proc. Roy. Soc. London, A193, pp. 120-146.
Poeter E.P. and Hill, M.C., 1996. Unrealistic parameter estimates
in inverse modeling: a problem or a benefit for model
calibration? Proceedings ModelCARE Conf., p. 227.

9.8
References

Poeter E.P. and Hill, M.C., 1997. Inverse models: A necessary next
step in ground-water modeling, Ground Water 35(2), 250-
260.
Poeter, E.P. and Anderson, D.R., 2005. Multi-model ranking and
inference in groundwater modeling, Ground Water, 43, no.
4, p. 597-605.
Poeter, E.P., Hill, M.C., Banta, E.R. 2005. MMRI, A computer code
for Multi-Model Ranking and Inference, a JUPITER API
Application Code, U.S. Geological Survey, Water-Resources
Investigations report.
Rango, A. and Martinec, J., 1995. Revisiting the degree-day
method for snowmelt computations, Water Resources
Bulletin. Vol. 31, no. 4, pp. 657-669.
Refsgaard, J.C., 1997. Parametrisation, calibration and validation
of distributed hydrological models, Journal of Hydrology,
198: 69-97.
Renschler, C.S. 2003. Designing geo-spatial interfaces to scale
process models: The GeoWEPP approach. Hydrological
Processes 17, p.1005–1017.
Rewerts, C. C., and Engel, B. A., 1993. ANSWERS on GRASS:
Integration of a watershed simulation with a geographic
information system. Abstracts, Proc., 8th Annu. GRASS GIS
User’s Conf. and Exhibition, Conf. Agenda and Listing of
Abstracts.
Rosenthal, W. D., Srinivasan, R., and Arnold, J., 1993. A GIS-
watershed hydrology model link to evaluate water resources
of the lower Colorado river in Texas. Proc., Application of
Advanced Information Technol.
Sansò F., 1996. Elementi di teoria della probabilità, 1st edition,
Città Studi Edizioni, pp. 111-119.
Savabi, M. R., Flanagan, D. C., Hebel, B., and Engel, B. A., 1995.
Application of WEPP and GIS-GRASS to a small watershed in
Indiana.’’ J. Soil and Water Conservation, 50(50), 477–483.

9.9
References

Schmid, F.; Fraefel, M.; Hegg, C., 2004. Unwetterschäden in der


Schweiz 1972-2002: Verteilung, Ursachen, Entwicklung. -
Wasser Energ. Luft 96, 1/2: 21-28.
Schulla J. and Jasper, K., 1999. Model description WaSim, IHW-
ETH, Zurich
Schulla J., 1997. Hydrologische Modellierung von Flussgebieten
zur Abschätzung der Folgen von Klimaänderungen, PhD.
Thesis ETH, Zurich
Seo D.J., 1998. Real-time estimation of rainfall fields using radar
rainfall and rain gage data, J. Hydrol., Vol. 208, pp. 37-52.
Shuttleworth, W. J., 1988. Evaporation from Amazonian Rainforest,
Proceedings of the Royal Society of London. Series B,
Biological Sciences, Vol. 233, No. 1272 (Apr. 22, 1988) , pp.
321-346
Shuttleworth, W. J., 1992. Evaporation, Chapter 4 in Handbook of
Hydrology, Edited by D. R. Maidment, McGraw-Hill, New
York.
Singh, V.P. and Woolhiser, D.A., 2002. Mathematical Modeling of
Watershed hydrology, Journal of Hydrologic Engineering,
7:4 (270-292).
St-Onge, B.A. and Achaichia N., 2001. Measuring forest canopy
height using a combination of lidar and aerial photography
data, International Archives of Photogrammetry and Remote
Sensing, Volume XXXIV-3/W4 Annapolis, MD, 22-2r4 Oct.
2001.
Strahler, A.H.,Muller, J.-P., MODIS Science Team Members, 1999.
MODIS BRDF/Albedo Product: Algorithm Theoretical Basis
Document Version 5.0,
http://modis.gsfc.nasa.gov/data/atbd/atbd_mod09.pdf, 2006.
Tarboton, D. G. and C. H. Luce, 1996. Utah Energy Balance Snow
Accumulation and Melt Model (UEB): Computer model
technical description and user’s guide, Utah Water Research
Laboratory and USDA Forest Service Intermountain
Research Station.

9.10
References

Tarboton, D. G., R. Woods, C. Pearson, U. Shankar, R. Henderson


and R. Ibbitt, 1999. Distributed Hydrologic Modeling using
GIS and Topmodel, Presentation at 19th Annual AGU
Hydrology Days, Fort Collins, Colorado, August 16-20
Tarboton, D.G., 1997. A new method for the determination of flow
directions and upslope areas in grid digital elevation models.
Water Resources Research, 33(2): 309-319.
Thornthwaite, C.W., and Mather, J.R., 1955. The water balance,
Publ. in Climatology, 8(1), Lab. of Climatology, Centerton,
N.J.
Todini, E., 1988. Rainfall-runoff modeling – past, present and
future, Journal of Hydrology, 100, 341-352.
Todini, E., 2001. A Bayesian technique for conditioning radar
precipitation estimates to rain-gauge measurements, Hydrol.
Earth System Sci., Vol. 5, No. 2, pp. 187-199.
Todini, E., 2005. Multi-Sensor Precipitation Measurements,
Integration, Calibration and Flood Forecasting (MUSIC),
Final Report, EU Project EVK1-CT-2000-0058.
Todini, E., and Ciarapica, L., 2001, The TOPKAPI model, (chapter
12), in: Mathematical models of large watershed hydrology,
V.P. Singh et al. (Eds), Water Resources Publications,
Littleton, Colorado.
Wesseling, C.G., D. Karssenberg, W.P.A. van Deursen, and P.A.
Burrough., 1996. Integrating dynamic environmental models
in GIS: The development of a Dynamic Modelling language,
Transaction in GIS 1, 40-48.
Wigmosta, M. S., Vail, L. W., and Lettenmaier, D. P. A., 1994.
Distributed hydrology vegetation model for complex terrain,
Water Resources Research, 30(6), 1665-79
WMO, 1986. Intercomparison of Models of Snowmelt Runoff,
Operational Hydrology Report No. 23, World Meteorological
Organization - No. 646.

9.11
References

Wooding, R.A., 1965. A hydraulic model for the catchment-stream


problem: I. Kinematic-wave theory; II. Numerical solutions;
III. Comparison with runoff observations, Journal of
Hydrology, Volume 191, Number 1, April 1997, pp. 371-
371(1)
Xu, C.Y. and Chen, D., 2005. Comparison of seven models for
estimation of evapotranspiration and groundwater recharge
usinglysimeter measurement data in German, Hydrol.
Process. 19, pp. 3717–3734
Y. Knyazikhin, J. Glassy, J. L. Privette, Y. Tian, A. Lotsch, Y. Zhang,
Y. Wang, J. T. Morisette, P.Votava, R.B. Myneni, R. R.
Nemani, S. W. Running, MODIS Leaf Area Index (LAI) and
Fraction of Photosynthetically Active Radiation Absorbed by
Vegetation (FPAR) Product (MOD15) Algorithm Theoretical
Basis Document,
http://eospso.gsfc.nasa.gov/atbd/modistables.html, 1999.
Ye, Z., Maidment, D.R., McKinney, D.C., 1996. Map-based surface
and subsurface flow simulation models: an object-oriented
and GIS approach, University of Texas, Austin.
Zollmann, C., 2000. Nachhaltige Bewässerungslandwirtschaft in
semi-ariden Gebieten, Entwicklung einer Upscaling technik
für die Hydrologische Modellierung, PhD. Thesis, ETH,
Zurich.

9.12
Appendix: Meteorological Network Stations Database

Appendix: Meteorological Network


Stations Database
In this appendix a description of the Meteorological Network
Stations Database (MNSD) tables is described.

Measurement_points
It archives the spatial component of the climatic stations, it has the
following attributes:
 feature_id – the station identification number stored as long

integer and set as primary key;


 east, north, elev – the coordinates stored as integer values;

 decr – the description of the station, stored as varchar(40).

Measurements
It stores the observed values, it has the following attributes:
 feature_id – the station identification number stored as long
integer and set as primary key and as foreign key;
 param_id – the parameters identification number stored as
integer and set as primary key and foreign key;
 meas_time – the time of the observation, stored as timestamp;

 meas_value – the observed value, stored as real;

Parameters
It is the look-up table to associate informations with different
parameters type, it has the following attributes:
 param_id – the parameters identification number stored as
integer and set as primary key;
 name – the name of the observation type stored as varchar(5);

 descr – the description for the observation type stored as


varchar(20);
 unit – the unit of the observation type stored as varchar(10);

A.1
Appendix: Meteorological Network Stations Database

Instruments
It stores information about instruments installed at different
measure points, it has the following attributes:
 instr_id – the unique identifier for each instrument stored as long

integer and set as primary key;


 feature_id – the station identification number stored as long

integer and set as foreign key;


 owner_id – the identification number of the organization owner of

the instrument stored as integer and set as foreign key;


 type_id – the identification number of the instrument type stored

as integer and set as foreign key;


 serial – the serial number of the instrument stored as varchar(30);

Instr_type
it is the look up table to relate instrument identifier with its
informations, it has the following attributes:
 type_id – the identification number of the instrument type stored

as integer and set as primary key;


 brand – the name of the brand of the instrument stored as

varchar(20);
 version – the version of the instrument stored as varchar(20);

Organizations
It stores the owner of the different instruments, it has the following
attributes:
 owner_id – the identification number of the organization owner of

the instrument, stored as integer and set as primary key;


 name – the name of the organization owing the instruments
stored as varchar(20) and set as primary key;
 address – the address of the organization stored as varchar(50)

 town – the town of the organization stored as varchar(15);

 nation – the nationality of the organization stored as varchar(15);

 phone – the phone number of the organization stored as


varchar(20);

A.2
Appendix: Meteorological Network Stations Database

 email – the contact email address stored as varchar(30);

 contact – the identifier name of the person of contact stored as

varchar(30);

Maintenance
It archives the maintenance operations done on the instruments, it
has the following attributes:
 main_id – the unique identifier of each operation stored as long

integer and set as primary key;


 main_time – the date and time of the operation stored as

timestamp;
 descr – the description of the executed operation stored as

varchar(50);
 instr_id – the identifier of the instrument stored as long integer

and set as foreign key;


 person_id – the name of the person that executed the operation

stored as long integer and set as a foreign key;

Persons
It store information about the person involved in the technical
maintenance of the network, it has the following attributes:
 pers_id – the unique identifier of the person involved in the

network maintenance, stored as long integer and set as primary


key;
 name – first name of the person stored as varchar(20);

 last – family name of the person stored as varchar(20);

 phone – the phone number of the person stored as varchar(20);

 email – e-mail of the person stored as varchar(30);

A.3